[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7EPHSN17SV2JVTNKKEHWHVF":3,"skill-same:01M7EPHSN17SV2JVTNKKEHWHVF":68},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":36,"commitSha":37,"skillDirectory":38,"forksCount":39,"repositoryHtmlUrl":40,"repositoryHomepageUrl":6,"ownerHtmlUrl":41,"ownerBlogUrl":42,"ownerDescription":6,"updatedAt":43,"install":44,"resources":57},"01M7EPHSN17SV2JVTNKKEHWHVF","multimodal-audiocraft",null,"PyTorch library for audio generation including text-to-music (MusicGen) and text-to-sound (AudioGen). Use when you need to generate music from text descriptions, create sound effects, or perform melody-conditioned music generation.","TanNhatCMS\u002Fjarvis","TanNhatCMS","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F113796420?v=4","01M753MV17KG5CD3152G3WQEQY","KietNT","01M7EKEDYVYQ6PWVJJMZ911VMX",0,"MIT","https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fams-imp-multimodal-audiocraft\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: multimodal-audiocraft\ndescription: PyTorch library for audio generation including text-to-music (MusicGen) and text-to-sound (AudioGen). Use when you need to generate music from text descriptions, create sound effects, or perform melody-conditioned music generation.\nversion: 1.0.0\nauthor: Orchestra Research\nlicense: MIT\ntags: [Multimodal, Audio Generation, Text-to-Music, Text-to-Audio, MusicGen]\ndependencies: [audiocraft, torch>=2.0.0, transformers>=4.30.0]\n---\n\n# AudioCraft: Audio Generation\n\nComprehensive guide to using Meta's AudioCraft for text-to-music and text-to-audio generation with MusicGen, AudioGen, and EnCodec.\n\n## When to use AudioCraft\n\n**Use AudioCraft when:**\n- Need to generate music from text descriptions\n- Creating sound effects and environmental audio\n- Building music generation applications\n- Need melody-conditioned music generation\n- Want stereo audio output\n- Require controllable music generation with style transfer\n\n**Key features:**\n- **MusicGen**: Text-to-music generation with melody conditioning\n- **AudioGen**: Text-to-sound effects generation\n- **EnCodec**: High-fidelity neural audio codec\n- **Multiple model sizes**: Small (300M) to Large (3.3B)\n- **Stereo support**: Full stereo audio generation\n- **Style conditioning**: MusicGen-Style for reference-based generation\n\n**Use alternatives instead:**\n- **Stable Audio**: For longer commercial music generation\n- **Bark**: For text-to-speech with music\u002Fsound effects\n- **Riffusion**: For spectogram-based music generation\n- **OpenAI Jukebox**: For raw audio generation with lyrics\n\n## Quick start\n\n### Installation\n\n```bash\n# From PyPI\npip install audiocraft\n\n# From GitHub (latest)\npip install git+https:\u002F\u002Fgithub.com\u002Ffacebookresearch\u002Faudiocraft.git\n\n# Or use HuggingFace Transformers\npip install transformers torch torchaudio\n```\n\n### Basic text-to-music (AudioCraft)\n\n```python\nimport torchaudio\nfrom audiocraft.models import MusicGen\n\n# Load model\nmodel = MusicGen.get_pretrained('facebook\u002Fmusicgen-small')\n\n# Set generation parameters\nmodel.set_generation_params(\n    duration=8,  # seconds\n    top_k=250,\n    temperature=1.0\n)\n\n# Generate from text\ndescriptions = [\"happy upbeat electronic dance music with synths\"]\nwav = model.generate(descriptions)\n\n# Save audio\ntorchaudio.save(\"output.wav\", wav[0].cpu(), sample_rate=32000)\n```\n\n### Using HuggingFace Transformers\n\n```python\nfrom transformers import AutoProcessor, MusicgenForConditionalGeneration\nimport scipy\n\n# Load model and processor\nprocessor = AutoProcessor.from_pretrained(\"facebook\u002Fmusicgen-small\")\nmodel = MusicgenForConditionalGeneration.from_pretrained(\"facebook\u002Fmusicgen-small\")\nmodel.to(\"cuda\")\n\n# Generate music\ninputs = processor(\n    text=[\"80s pop track with bassy drums and synth\"],\n    padding=True,\n    return_tensors=\"pt\"\n).to(\"cuda\")\n\naudio_values = model.generate(\n    **inputs,\n    do_sample=True,\n    guidance_scale=3,\n    max_new_tokens=256\n)\n\n# Save\nsampling_rate = model.config.audio_encoder.sampling_rate\nscipy.io.wavfile.write(\"output.wav\", rate=sampling_rate, data=audio_values[0, 0].cpu().numpy())\n```\n\n### Text-to-sound with AudioGen\n\n```python\nfrom audiocraft.models import AudioGen\n\n# Load AudioGen\nmodel = AudioGen.get_pretrained('facebook\u002Faudiogen-medium')\n\nmodel.set_generation_params(duration=5)\n\n# Generate sound effects\ndescriptions = [\"dog barking in a park with birds chirping\"]\nwav = model.generate(descriptions)\n\ntorchaudio.save(\"sound.wav\", wav[0].cpu(), sample_rate=16000)\n```\n\n## Core concepts\n\n### Architecture overview\n\n```\nAudioCraft Architecture:\n┌──────────────────────────────────────────────────────────────┐\n│                    Text Encoder (T5)                          │\n│                         │                                     │\n│                    Text Embeddings                            │\n└────────────────────────┬─────────────────────────────────────┘\n                         │\n┌────────────────────────▼─────────────────────────────────────┐\n│              Transformer Decoder (LM)                         │\n│     Auto-regressively generates audio tokens                  │\n│     Using efficient token interleaving patterns               │\n└────────────────────────┬─────────────────────────────────────┘\n                         │\n┌────────────────────────▼─────────────────────────────────────┐\n│                EnCodec Audio Decoder                          │\n│        Converts tokens back to audio waveform                 │\n└──────────────────────────────────────────────────────────────┘\n```\n\n### Model variants\n\n| Model | Size | Description | Use Case |\n|-------|------|-------------|----------|\n| `musicgen-small` | 300M | Text-to-music | Quick generation |\n| `musicgen-medium` | 1.5B | Text-to-music | Balanced |\n| `musicgen-large` | 3.3B | Text-to-music | Best quality |\n| `musicgen-melody` | 1.5B | Text + melody | Melody conditioning |\n| `musicgen-melody-large` | 3.3B | Text + melody | Best melody |\n| `musicgen-stereo-*` | Varies | Stereo output | Stereo generation |\n| `musicgen-style` | 1.5B | Style transfer | Reference-based |\n| `audiogen-medium` | 1.5B | Text-to-sound | Sound effects |\n\n### Generation parameters\n\n| Parameter | Default | Description |\n|-----------|---------|-------------|\n| `duration` | 8.0 | Length in seconds (1-120) |\n| `top_k` | 250 | Top-k sampling |\n| `top_p` | 0.0 | Nucleus sampling (0 = disabled) |\n| `temperature` | 1.0 | Sampling temperature |\n| `cfg_coef` | 3.0 | Classifier-free guidance |\n\n## MusicGen usage\n\n### Text-to-music generation\n\n```python\nfrom audiocraft.models import MusicGen\nimport torchaudio\n\nmodel = MusicGen.get_pretrained('facebook\u002Fmusicgen-medium')\n\n# Configure generation\nmodel.set_generation_params(\n    duration=30,          # Up to 30 seconds\n    top_k=250,            # Sampling diversity\n    top_p=0.0,            # 0 = use top_k only\n    temperature=1.0,      # Creativity (higher = more varied)\n    cfg_coef=3.0          # Text adherence (higher = stricter)\n)\n\n# Generate multiple samples\ndescriptions = [\n    \"epic orchestral soundtrack with strings and brass\",\n    \"chill lo-fi hip hop beat with jazzy piano\",\n    \"energetic rock song with electric guitar\"\n]\n\n# Generate (returns [batch, channels, samples])\nwav = model.generate(descriptions)\n\n# Save each\nfor i, audio in enumerate(wav):\n    torchaudio.save(f\"music_{i}.wav\", audio.cpu(), sample_rate=32000)\n```\n\n### Melody-conditioned generation\n\n```python\nfrom audiocraft.models import MusicGen\nimport torchaudio\n\n# Load melody model\nmodel = MusicGen.get_pretrained('facebook\u002Fmusicgen-melody')\nmodel.set_generation_params(duration=30)\n\n# Load melody audio\nmelody, sr = torchaudio.load(\"melody.wav\")\n\n# Generate with melody conditioning\ndescriptions = [\"acoustic guitar folk song\"]\nwav = model.generate_with_chroma(descriptions, melody, sr)\n\ntorchaudio.save(\"melody_conditioned.wav\", wav[0].cpu(), sample_rate=32000)\n```\n\n### Stereo generation\n\n```python\nfrom audiocraft.models import MusicGen\n\n# Load stereo model\nmodel = MusicGen.get_pretrained('facebook\u002Fmusicgen-stereo-medium')\nmodel.set_generation_params(duration=15)\n\ndescriptions = [\"ambient electronic music with wide stereo panning\"]\nwav = model.generate(descriptions)\n\n# wav shape: [batch, 2, samples] for stereo\nprint(f\"Stereo shape: {wav.shape}\")  # [1, 2, 480000]\ntorchaudio.save(\"stereo.wav\", wav[0].cpu(), sample_rate=32000)\n```\n\n### Audio continuation\n\n```python\nfrom transformers import AutoProcessor, MusicgenForConditionalGeneration\n\nprocessor = AutoProcessor.from_pretrained(\"facebook\u002Fmusicgen-medium\")\nmodel = MusicgenForConditionalGeneration.from_pretrained(\"facebook\u002Fmusicgen-medium\")\n\n# Load audio to continue\nimport torchaudio\naudio, sr = torchaudio.load(\"intro.wav\")\n\n# Process with text and audio\ninputs = processor(\n    audio=audio.squeeze().numpy(),\n    sampling_rate=sr,\n    text=[\"continue with a epic chorus\"],\n    padding=True,\n    return_tensors=\"pt\"\n)\n\n# Generate continuation\naudio_values = model.generate(**inputs, do_sample=True, guidance_scale=3, max_new_tokens=512)\n```\n\n## MusicGen-Style usage\n\n### Style-conditioned generation\n\n```python\nfrom audiocraft.models import MusicGen\n\n# Load style model\nmodel = MusicGen.get_pretrained('facebook\u002Fmusicgen-style')\n\n# Configure generation with style\nmodel.set_generation_params(\n    duration=30,\n    cfg_coef=3.0,\n    cfg_coef_beta=5.0  # Style influence\n)\n\n# Configure style conditioner\nmodel.set_style_conditioner_params(\n    eval_q=3,          # RVQ quantizers (1-6)\n    excerpt_length=3.0  # Style excerpt length\n)\n\n# Load style reference\nstyle_audio, sr = torchaudio.load(\"reference_style.wav\")\n\n# Generate with text + style\ndescriptions = [\"upbeat dance track\"]\nwav = model.generate_with_style(descriptions, style_audio, sr)\n```\n\n### Style-only generation (no text)\n\n```python\n# Generate matching style without text prompt\nmodel.set_generation_params(\n    duration=30,\n    cfg_coef=3.0,\n    cfg_coef_beta=None  # Disable double CFG for style-only\n)\n\nwav = model.generate_with_style([None], style_audio, sr)\n```\n\n## AudioGen usage\n\n### Sound effect generation\n\n```python\nfrom audiocraft.models import AudioGen\nimport torchaudio\n\nmodel = AudioGen.get_pretrained('facebook\u002Faudiogen-medium')\nmodel.set_generation_params(duration=10)\n\n# Generate various sounds\ndescriptions = [\n    \"thunderstorm with heavy rain and lightning\",\n    \"busy city traffic with car horns\",\n    \"ocean waves crashing on rocks\",\n    \"crackling campfire in forest\"\n]\n\nwav = model.generate(descriptions)\n\nfor i, audio in enumerate(wav):\n    torchaudio.save(f\"sound_{i}.wav\", audio.cpu(), sample_rate=16000)\n```\n\n## EnCodec usage\n\n### Audio compression\n\n```python\nfrom audiocraft.models import CompressionModel\nimport torch\nimport torchaudio\n\n# Load EnCodec\nmodel = CompressionModel.get_pretrained('facebook\u002Fencodec_32khz')\n\n# Load audio\nwav, sr = torchaudio.load(\"audio.wav\")\n\n# Ensure correct sample rate\nif sr != 32000:\n    resampler = torchaudio.transforms.Resample(sr, 32000)\n    wav = resampler(wav)\n\n# Encode to tokens\nwith torch.no_grad():\n    encoded = model.encode(wav.unsqueeze(0))\n    codes = encoded[0]  # Audio codes\n\n# Decode back to audio\nwith torch.no_grad():\n    decoded = model.decode(codes)\n\ntorchaudio.save(\"reconstructed.wav\", decoded[0].cpu(), sample_rate=32000)\n```\n\n## Common workflows\n\n### Workflow 1: Music generation pipeline\n\n```python\nimport torch\nimport torchaudio\nfrom audiocraft.models import MusicGen\n\nclass MusicGenerator:\n    def __init__(self, model_name=\"facebook\u002Fmusicgen-medium\"):\n        self.model = MusicGen.get_pretrained(model_name)\n        self.sample_rate = 32000\n\n    def generate(self, prompt, duration=30, temperature=1.0, cfg=3.0):\n        self.model.set_generation_params(\n            duration=duration,\n            top_k=250,\n            temperature=temperature,\n            cfg_coef=cfg\n        )\n\n        with torch.no_grad():\n            wav = self.model.generate([prompt])\n\n        return wav[0].cpu()\n\n    def generate_batch(self, prompts, duration=30):\n        self.model.set_generation_params(duration=duration)\n\n        with torch.no_grad():\n            wav = self.model.generate(prompts)\n\n        return wav.cpu()\n\n    def save(self, audio, path):\n        torchaudio.save(path, audio, sample_rate=self.sample_rate)\n\n# Usage\ngenerator = MusicGenerator()\naudio = generator.generate(\n    \"epic cinematic orchestral music\",\n    duration=30,\n    temperature=1.0\n)\ngenerator.save(audio, \"epic_music.wav\")\n```\n\n### Workflow 2: Sound design batch processing\n\n```python\nimport json\nfrom pathlib import Path\nfrom audiocraft.models import AudioGen\nimport torchaudio\n\ndef batch_generate_sounds(sound_specs, output_dir):\n    \"\"\"\n    Generate multiple sounds from specifications.\n\n    Args:\n        sound_specs: list of {\"name\": str, \"description\": str, \"duration\": float}\n        output_dir: output directory path\n    \"\"\"\n    model = AudioGen.get_pretrained('facebook\u002Faudiogen-medium')\n    output_dir = Path(output_dir)\n    output_dir.mkdir(exist_ok=True)\n\n    results = []\n\n    for spec in sound_specs:\n        model.set_generation_params(duration=spec.get(\"duration\", 5))\n\n        wav = model.generate([spec[\"description\"]])\n\n        output_path = output_dir \u002F f\"{spec['name']}.wav\"\n        torchaudio.save(str(output_path), wav[0].cpu(), sample_rate=16000)\n\n        results.append({\n            \"name\": spec[\"name\"],\n            \"path\": str(output_path),\n            \"description\": spec[\"description\"]\n        })\n\n    return results\n\n# Usage\nsounds = [\n    {\"name\": \"explosion\", \"description\": \"massive explosion with debris\", \"duration\": 3},\n    {\"name\": \"footsteps\", \"description\": \"footsteps on wooden floor\", \"duration\": 5},\n    {\"name\": \"door\", \"description\": \"wooden door creaking and closing\", \"duration\": 2}\n]\n\nresults = batch_generate_sounds(sounds, \"sound_effects\u002F\")\n```\n\n### Workflow 3: Gradio demo\n\n```python\nimport gradio as gr\nimport torch\nimport torchaudio\nfrom audiocraft.models import MusicGen\n\nmodel = MusicGen.get_pretrained('facebook\u002Fmusicgen-small')\n\ndef generate_music(prompt, duration, temperature, cfg_coef):\n    model.set_generation_params(\n        duration=duration,\n        temperature=temperature,\n        cfg_coef=cfg_coef\n    )\n\n    with torch.no_grad():\n        wav = model.generate([prompt])\n\n    # Save to temp file\n    path = \"temp_output.wav\"\n    torchaudio.save(path, wav[0].cpu(), sample_rate=32000)\n    return path\n\ndemo = gr.Interface(\n    fn=generate_music,\n    inputs=[\n        gr.Textbox(label=\"Music Description\", placeholder=\"upbeat electronic dance music\"),\n        gr.Slider(1, 30, value=8, label=\"Duration (seconds)\"),\n        gr.Slider(0.5, 2.0, value=1.0, label=\"Temperature\"),\n        gr.Slider(1.0, 10.0, value=3.0, label=\"CFG Coefficient\")\n    ],\n    outputs=gr.Audio(label=\"Generated Music\"),\n    title=\"MusicGen Demo\"\n)\n\ndemo.launch()\n```\n\n## Performance optimization\n\n### Memory optimization\n\n```python\n# Use smaller model\nmodel = MusicGen.get_pretrained('facebook\u002Fmusicgen-small')\n\n# Clear cache between generations\ntorch.cuda.empty_cache()\n\n# Generate shorter durations\nmodel.set_generation_params(duration=10)  # Instead of 30\n\n# Use half precision\nmodel = model.half()\n```\n\n### Batch processing efficiency\n\n```python\n# Process multiple prompts at once (more efficient)\ndescriptions = [\"prompt1\", \"prompt2\", \"prompt3\", \"prompt4\"]\nwav = model.generate(descriptions)  # Single batch\n\n# Instead of\nfor desc in descriptions:\n    wav = model.generate([desc])  # Multiple batches (slower)\n```\n\n### GPU memory requirements\n\n| Model | FP32 VRAM | FP16 VRAM |\n|-------|-----------|-----------|\n| musicgen-small | ~4GB | ~2GB |\n| musicgen-medium | ~8GB | ~4GB |\n| musicgen-large | ~16GB | ~8GB |\n\n## Common issues\n\n| Issue | Solution |\n|-------|----------|\n| CUDA OOM | Use smaller model, reduce duration |\n| Poor quality | Increase cfg_coef, better prompts |\n| Generation too short | Check max duration setting |\n| Audio artifacts | Try different temperature |\n| Stereo not working | Use stereo model variant |\n\n## References\n\n- **[Advanced Usage](references\u002Fadvanced-usage.md)** - Training, fine-tuning, deployment\n- **[Troubleshooting](references\u002Ftroubleshooting.md)** - Common issues and solutions\n\n## Resources\n\n- **GitHub**: https:\u002F\u002Fgithub.com\u002Ffacebookresearch\u002Faudiocraft\n- **Paper (MusicGen)**: https:\u002F\u002Farxiv.org\u002Fabs\u002F2306.05284\n- **Paper (AudioGen)**: https:\u002F\u002Farxiv.org\u002Fabs\u002F2209.15352\n- **HuggingFace**: https:\u002F\u002Fhuggingface.co\u002Ffacebook\u002Fmusicgen-small\n- **Demo**: https:\u002F\u002Fhuggingface.co\u002Fspaces\u002Ffacebook\u002FMusicGen\n","# AudioCraft: Audio Generation\n\nComprehensive guide to using Meta's AudioCraft for text-to-music and text-to-audio generation with MusicGen, AudioGen, and EnCodec.\n\n## When to use AudioCraft\n\n**Use AudioCraft when:**\n- Need to generate music from text descriptions\n- Creating sound effects and environmental audio\n- Building music generation applications\n- Need melody-conditioned music generation\n- Want stereo audio output\n- Require controllable music generation with style transfer\n\n**Key features:**\n- **MusicGen**: Text-to-music generation with melody conditioning\n- **AudioGen**: Text-to-sound effects generation\n- **EnCodec**: High-fidelity neural audio codec\n- **Multiple model sizes**: Small (300M) to Large (3.3B)\n- **Stereo support**: Full stereo audio generation\n- **Style conditioning**: MusicGen-Style for reference-based generation\n\n**Use alternatives instead:**\n- **Stable Audio**: For longer commercial music generation\n- **Bark**: For text-to-speech with music\u002Fsound effects\n- **Riffusion**: For spectogram-based music generation\n- **OpenAI Jukebox**: For raw audio generation with lyrics\n\n## Quick start\n\n### Installation\n\n```bash\n# From PyPI\npip install audiocraft\n\n# From GitHub (latest)\npip install git+https:\u002F\u002Fgithub.com\u002Ffacebookresearch\u002Faudiocraft.git\n\n# Or use HuggingFace Transformers\npip install transformers torch torchaudio\n```\n\n### Basic text-to-music (AudioCraft)\n\n```python\nimport torchaudio\nfrom audiocraft.models import MusicGen\n\n# Load model\nmodel = MusicGen.get_pretrained('facebook\u002Fmusicgen-small')\n\n# Set generation parameters\nmodel.set_generation_params(\n    duration=8,  # seconds\n    top_k=250,\n    temperature=1.0\n)\n\n# Generate from text\ndescriptions = [\"happy upbeat electronic dance music with synths\"]\nwav = model.generate(descriptions)\n\n# Save audio\ntorchaudio.save(\"output.wav\", wav[0].cpu(), sample_rate=32000)\n```\n\n### Using HuggingFace Transformers\n\n```python\nfrom transformers import AutoProcessor, MusicgenForConditionalGeneration\nimport scipy\n\n# Load model and processor\nprocessor = AutoProcessor.from_pretrained(\"facebook\u002Fmusicgen-small\")\nmodel = MusicgenForConditionalGeneration.from_pretrained(\"facebook\u002Fmusicgen-small\")\nmodel.to(\"cuda\")\n\n# Generate music\ninputs = processor(\n    text=[\"80s pop track with bassy drums and synth\"],\n    padding=True,\n    return_tensors=\"pt\"\n).to(\"cuda\")\n\naudio_values = model.generate(\n    **inputs,\n    do_sample=True,\n    guidance_scale=3,\n    max_new_tokens=256\n)\n\n# Save\nsampling_rate = model.config.audio_encoder.sampling_rate\nscipy.io.wavfile.write(\"output.wav\", rate=sampling_rate, data=audio_values[0, 0].cpu().numpy())\n```\n\n### Text-to-sound with AudioGen\n\n```python\nfrom audiocraft.models import AudioGen\n\n# Load AudioGen\nmodel = AudioGen.get_pretrained('facebook\u002Faudiogen-medium')\n\nmodel.set_generation_params(duration=5)\n\n# Generate sound effects\ndescriptions = [\"dog barking in a park with birds chirping\"]\nwav = model.generate(descriptions)\n\ntorchaudio.save(\"sound.wav\", wav[0].cpu(), sample_rate=16000)\n```\n\n## Core concepts\n\n### Architecture overview\n\n```\nAudioCraft Architecture:\n┌──────────────────────────────────────────────────────────────┐\n│                    Text Encoder (T5)                          │\n│                         │                                     │\n│                    Text Embeddings                            │\n└────────────────────────┬─────────────────────────────────────┘\n                         │\n┌────────────────────────▼─────────────────────────────────────┐\n│              Transformer Decoder (LM)                         │\n│     Auto-regressively generates audio tokens                  │\n│     Using efficient token interleaving patterns               │\n└────────────────────────┬─────────────────────────────────────┘\n                         │\n┌────────────────────────▼─────────────────────────────────────┐\n│                EnCodec Audio Decoder                          │\n│        Converts tokens back to audio waveform                 │\n└──────────────────────────────────────────────────────────────┘\n```\n\n### Model variants\n\n| Model | Size | Description | Use Case |\n|-------|------|-------------|----------|\n| `musicgen-small` | 300M | Text-to-music | Quick generation |\n| `musicgen-medium` | 1.5B | Text-to-music | Balanced |\n| `musicgen-large` | 3.3B | Text-to-music | Best quality |\n| `musicgen-melody` | 1.5B | Text + melody | Melody conditioning |\n| `musicgen-melody-large` | 3.3B | Text + melody | Best melody |\n| `musicgen-stereo-*` | Varies | Stereo output | Stereo generation |\n| `musicgen-style` | 1.5B | Style transfer | Reference-based |\n| `audiogen-medium` | 1.5B | Text-to-sound | Sound effects |\n\n### Generation parameters\n\n| Parameter | Default | Description |\n|-----------|---------|-------------|\n| `duration` | 8.0 | Length in seconds (1-120) |\n| `top_k` | 250 | Top-k sampling |\n| `top_p` | 0.0 | Nucleus sampling (0 = disabled) |\n| `temperature` | 1.0 | Sampling temperature |\n| `cfg_coef` | 3.0 | Classifier-free guidance |\n\n## MusicGen usage\n\n### Text-to-music generation\n\n```python\nfrom audiocraft.models import MusicGen\nimport torchaudio\n\nmodel = MusicGen.get_pretrained('facebook\u002Fmusicgen-medium')\n\n# Configure generation\nmodel.set_generation_params(\n    duration=30,          # Up to 30 seconds\n    top_k=250,            # Sampling diversity\n    top_p=0.0,            # 0 = use top_k only\n    temperature=1.0,      # Creativity (higher = more varied)\n    cfg_coef=3.0          # Text adherence (higher = stricter)\n)\n\n# Generate multiple samples\ndescriptions = [\n    \"epic orchestral soundtrack with strings and brass\",\n    \"chill lo-fi hip hop beat with jazzy piano\",\n    \"energetic rock song with electric guitar\"\n]\n\n# Generate (returns [batch, channels, samples])\nwav = model.generate(descriptions)\n\n# Save each\nfor i, audio in enumerate(wav):\n    torchaudio.save(f\"music_{i}.wav\", audio.cpu(), sample_rate=32000)\n```\n\n### Melody-conditioned generation\n\n```python\nfrom audiocraft.models import MusicGen\nimport torchaudio\n\n# Load melody model\nmodel = MusicGen.get_pretrained('facebook\u002Fmusicgen-melody')\nmodel.set_generation_params(duration=30)\n\n# Load melody audio\nmelody, sr = torchaudio.load(\"melody.wav\")\n\n# Generate with melody conditioning\ndescriptions = [\"acoustic guitar folk song\"]\nwav = model.generate_with_chroma(descriptions, melody, sr)\n\ntorchaudio.save(\"melody_conditioned.wav\", wav[0].cpu(), sample_rate=32000)\n```\n\n### Stereo generation\n\n```python\nfrom audiocraft.models import MusicGen\n\n# Load stereo model\nmodel = MusicGen.get_pretrained('facebook\u002Fmusicgen-stereo-medium')\nmodel.set_generation_params(duration=15)\n\ndescriptions = [\"ambient electronic music with wide stereo panning\"]\nwav = model.generate(descriptions)\n\n# wav shape: [batch, 2, samples] for stereo\nprint(f\"Stereo shape: {wav.shape}\")  # [1, 2, 480000]\ntorchaudio.save(\"stereo.wav\", wav[0].cpu(), sample_rate=32000)\n```\n\n### Audio continuation\n\n```python\nfrom transformers import AutoProcessor, MusicgenForConditionalGeneration\n\nprocessor = AutoProcessor.from_pretrained(\"facebook\u002Fmusicgen-medium\")\nmodel = MusicgenForConditionalGeneration.from_pretrained(\"facebook\u002Fmusicgen-medium\")\n\n# Load audio to continue\nimport torchaudio\naudio, sr = torchaudio.load(\"intro.wav\")\n\n# Process with text and audio\ninputs = processor(\n    audio=audio.squeeze().numpy(),\n    sampling_rate=sr,\n    text=[\"continue with a epic chorus\"],\n    padding=True,\n    return_tensors=\"pt\"\n)\n\n# Generate continuation\naudio_values = model.generate(**inputs, do_sample=True, guidance_scale=3, max_new_tokens=512)\n```\n\n## MusicGen-Style usage\n\n### Style-conditioned generation\n\n```python\nfrom audiocraft.models import MusicGen\n\n# Load style model\nmodel = MusicGen.get_pretrained('facebook\u002Fmusicgen-style')\n\n# Configure generation with style\nmodel.set_generation_params(\n    duration=30,\n    cfg_coef=3.0,\n    cfg_coef_beta=5.0  # Style influence\n)\n\n# Configure style conditioner\nmodel.set_style_conditioner_params(\n    eval_q=3,          # RVQ quantizers (1-6)\n    excerpt_length=3.0  # Style excerpt length\n)\n\n# Load style reference\nstyle_audio, sr = torchaudio.load(\"reference_style.wav\")\n\n# Generate with text + style\ndescriptions = [\"upbeat dance track\"]\nwav = model.generate_with_style(descriptions, style_audio, sr)\n```\n\n### Style-only generation (no text)\n\n```python\n# Generate matching style without text prompt\nmodel.set_generation_params(\n    duration=30,\n    cfg_coef=3.0,\n    cfg_coef_beta=None  # Disable double CFG for style-only\n)\n\nwav = model.generate_with_style([None], style_audio, sr)\n```\n\n## AudioGen usage\n\n### Sound effect generation\n\n```python\nfrom audiocraft.models import AudioGen\nimport torchaudio\n\nmodel = AudioGen.get_pretrained('facebook\u002Faudiogen-medium')\nmodel.set_generation_params(duration=10)\n\n# Generate various sounds\ndescriptions = [\n    \"thunderstorm with heavy rain and lightning\",\n    \"busy city traffic with car horns\",\n    \"ocean waves crashing on rocks\",\n    \"crackling campfire in forest\"\n]\n\nwav = model.generate(descriptions)\n\nfor i, audio in enumerate(wav):\n    torchaudio.save(f\"sound_{i}.wav\", audio.cpu(), sample_rate=16000)\n```\n\n## EnCodec usage\n\n### Audio compression\n\n```python\nfrom audiocraft.models import CompressionModel\nimport torch\nimport torchaudio\n\n# Load EnCodec\nmodel = CompressionModel.get_pretrained('facebook\u002Fencodec_32khz')\n\n# Load audio\nwav, sr = torchaudio.load(\"audio.wav\")\n\n# Ensure correct sample rate\nif sr != 32000:\n    resampler = torchaudio.transforms.Resample(sr, 32000)\n    wav = resampler(wav)\n\n# Encode to tokens\nwith torch.no_grad():\n    encoded = model.encode(wav.unsqueeze(0))\n    codes = encoded[0]  # Audio codes\n\n# Decode back to audio\nwith torch.no_grad():\n    decoded = model.decode(codes)\n\ntorchaudio.save(\"reconstructed.wav\", decoded[0].cpu(), sample_rate=32000)\n```\n\n## Common workflows\n\n### Workflow 1: Music generation pipeline\n\n```python\nimport torch\nimport torchaudio\nfrom audiocraft.models import MusicGen\n\nclass MusicGenerator:\n    def __init__(self, model_name=\"facebook\u002Fmusicgen-medium\"):\n        self.model = MusicGen.get_pretrained(model_name)\n        self.sample_rate = 32000\n\n    def generate(self, prompt, duration=30, temperature=1.0, cfg=3.0):\n        self.model.set_generation_params(\n            duration=duration,\n            top_k=250,\n            temperature=temperature,\n            cfg_coef=cfg\n        )\n\n        with torch.no_grad():\n            wav = self.model.generate([prompt])\n\n        return wav[0].cpu()\n\n    def generate_batch(self, prompts, duration=30):\n        self.model.set_generation_params(duration=duration)\n\n        with torch.no_grad():\n            wav = self.model.generate(prompts)\n\n        return wav.cpu()\n\n    def save(self, audio, path):\n        torchaudio.save(path, audio, sample_rate=self.sample_rate)\n\n# Usage\ngenerator = MusicGenerator()\naudio = generator.generate(\n    \"epic cinematic orchestral music\",\n    duration=30,\n    temperature=1.0\n)\ngenerator.save(audio, \"epic_music.wav\")\n```\n\n### Workflow 2: Sound design batch processing\n\n```python\nimport json\nfrom pathlib import Path\nfrom audiocraft.models import AudioGen\nimport torchaudio\n\ndef batch_generate_sounds(sound_specs, output_dir):\n    \"\"\"\n    Generate multiple sounds from specifications.\n\n    Args:\n        sound_specs: list of {\"name\": str, \"description\": str, \"duration\": float}\n        output_dir: output directory path\n    \"\"\"\n    model = AudioGen.get_pretrained('facebook\u002Faudiogen-medium')\n    output_dir = Path(output_dir)\n    output_dir.mkdir(exist_ok=True)\n\n    results = []\n\n    for spec in sound_specs:\n        model.set_generation_params(duration=spec.get(\"duration\", 5))\n\n        wav = model.generate([spec[\"description\"]])\n\n        output_path = output_dir \u002F f\"{spec['name']}.wav\"\n        torchaudio.save(str(output_path), wav[0].cpu(), sample_rate=16000)\n\n        results.append({\n            \"name\": spec[\"name\"],\n            \"path\": str(output_path),\n            \"description\": spec[\"description\"]\n        })\n\n    return results\n\n# Usage\nsounds = [\n    {\"name\": \"explosion\", \"description\": \"massive explosion with debris\", \"duration\": 3},\n    {\"name\": \"footsteps\", \"description\": \"footsteps on wooden floor\", \"duration\": 5},\n    {\"name\": \"door\", \"description\": \"wooden door creaking and closing\", \"duration\": 2}\n]\n\nresults = batch_generate_sounds(sounds, \"sound_effects\u002F\")\n```\n\n### Workflow 3: Gradio demo\n\n```python\nimport gradio as gr\nimport torch\nimport torchaudio\nfrom audiocraft.models import MusicGen\n\nmodel = MusicGen.get_pretrained('facebook\u002Fmusicgen-small')\n\ndef generate_music(prompt, duration, temperature, cfg_coef):\n    model.set_generation_params(\n        duration=duration,\n        temperature=temperature,\n        cfg_coef=cfg_coef\n    )\n\n    with torch.no_grad():\n        wav = model.generate([prompt])\n\n    # Save to temp file\n    path = \"temp_output.wav\"\n    torchaudio.save(path, wav[0].cpu(), sample_rate=32000)\n    return path\n\ndemo = gr.Interface(\n    fn=generate_music,\n    inputs=[\n        gr.Textbox(label=\"Music Description\", placeholder=\"upbeat electronic dance music\"),\n        gr.Slider(1, 30, value=8, label=\"Duration (seconds)\"),\n        gr.Slider(0.5, 2.0, value=1.0, label=\"Temperature\"),\n        gr.Slider(1.0, 10.0, value=3.0, label=\"CFG Coefficient\")\n    ],\n    outputs=gr.Audio(label=\"Generated Music\"),\n    title=\"MusicGen Demo\"\n)\n\ndemo.launch()\n```\n\n## Performance optimization\n\n### Memory optimization\n\n```python\n# Use smaller model\nmodel = MusicGen.get_pretrained('facebook\u002Fmusicgen-small')\n\n# Clear cache between generations\ntorch.cuda.empty_cache()\n\n# Generate shorter durations\nmodel.set_generation_params(duration=10)  # Instead of 30\n\n# Use half precision\nmodel = model.half()\n```\n\n### Batch processing efficiency\n\n```python\n# Process multiple prompts at once (more efficient)\ndescriptions = [\"prompt1\", \"prompt2\", \"prompt3\", \"prompt4\"]\nwav = model.generate(descriptions)  # Single batch\n\n# Instead of\nfor desc in descriptions:\n    wav = model.generate([desc])  # Multiple batches (slower)\n```\n\n### GPU memory requirements\n\n| Model | FP32 VRAM | FP16 VRAM |\n|-------|-----------|-----------|\n| musicgen-small | ~4GB | ~2GB |\n| musicgen-medium | ~8GB | ~4GB |\n| musicgen-large | ~16GB | ~8GB |\n\n## Common issues\n\n| Issue | Solution |\n|-------|----------|\n| CUDA OOM | Use smaller model, reduce duration |\n| Poor quality | Increase cfg_coef, better prompts |\n| Generation too short | Check max duration setting |\n| Audio artifacts | Try different temperature |\n| Stereo not working | Use stereo model variant |\n\n## References\n\n- **[Advanced Usage](references\u002Fadvanced-usage.md)** - Training, fine-tuning, deployment\n- **[Troubleshooting](references\u002Ftroubleshooting.md)** - Common issues and solutions\n\n## Resources\n\n- **GitHub**: https:\u002F\u002Fgithub.com\u002Ffacebookresearch\u002Faudiocraft\n- **Paper (MusicGen)**: https:\u002F\u002Farxiv.org\u002Fabs\u002F2306.05284\n- **Paper (AudioGen)**: https:\u002F\u002Farxiv.org\u002Fabs\u002F2209.15352\n- **HuggingFace**: https:\u002F\u002Fhuggingface.co\u002Ffacebook\u002Fmusicgen-small\n- **Demo**: https:\u002F\u002Fhuggingface.co\u002Fspaces\u002Ffacebook\u002FMusicGen\n",{"name":5,"tags":24,"author":30,"license":15,"version":31,"description":7,"dependencies":32},[25,26,27,28,29],"Multimodal","Audio Generation","Text-to-Music","Text-to-Audio","MusicGen","Orchestra Research","1.0.0",[33,34,35],"audiocraft","torch>=2.0.0","transformers>=4.30.0","main","af388a77c2fe5e0b4fd879cca4300a0ffa9a6579","skills\u002Fams-imp-multimodal-audiocraft",1,"https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis","https:\u002F\u002Fgithub.com\u002FTanNhatCMS","https:\u002F\u002Fmrkiet.name.vn","2026-11-07 14:23:08",{"source":8,"skillName":5,"commands":45,"prompt":55,"zipAvailable":56},[46,49,52],{"tool":47,"command":48},"npx","npx skills add TanNhatCMS\u002Fjarvis --skill multimodal-audiocraft",{"tool":50,"command":51},"bunx","bunx skills add TanNhatCMS\u002Fjarvis --skill multimodal-audiocraft",{"tool":53,"command":54},"pnpm","pnpm dlx skills add TanNhatCMS\u002Fjarvis --skill multimodal-audiocraft","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fams-imp-multimodal-audiocraft\u002FSKILL.md",true,[58,64],{"path":59,"resourceType":60,"resourceTypeDesc":61,"sizeBytes":62,"sourceHtmlUrl":63},"references\u002Fadvanced-usage.md","REFERENCE","参考文档",16953,"https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fams-imp-multimodal-audiocraft\u002Freferences\u002Fadvanced-usage.md",{"path":65,"resourceType":60,"resourceTypeDesc":61,"sizeBytes":66,"sourceHtmlUrl":67},"references\u002Ftroubleshooting.md",11412,"https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fams-imp-multimodal-audiocraft\u002Freferences\u002Ftroubleshooting.md",[69,76,84,91,98,105],{"skillNo":70,"name":71,"nameZh":6,"description":72,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":73,"categories":74,"keywords":75,"publishedAt":19},"01M7EWYJR0J8DARY7E59Z6Y9MV","web-scrape","Turn web pages into LLM-ready markdown via the Firecrawl API (renders JS, anti-bot handling). Use when browser_open\u002Ffetch fails on JS-heavy pages, or when you need clean markdown\u002Fstructured extraction from a site. Requires FIRECRAWL_API_KEY.","https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fweb-scrape\u002FSKILL.md",[],[],{"skillNo":77,"name":78,"nameZh":6,"description":79,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":80,"sourceHtmlUrl":81,"categories":82,"keywords":83,"publishedAt":19},"01M7EWYBN5P6F2AJG3FR24T9BP","design-taste-frontend","tasteskill v2, the anti-slop frontend skill. Art direction for websites, distilled from a study of 247 curated gallery sites (One Page Love, Awwwards Sites of the Day). Deals researched directions (idea, composition, type, palette, photographs, section blueprints) from a library, enforces a strict no-slop house style with a real-browser inspector, and runs a look-and-fix loop on screenshots. Use this whenever you build, design, redesign, restyle, polish or review a landing page, marketing site, portfolio, product page, one-pager, hero or any public-facing web page, in any stack (HTML\u002FCSS, React, Next.js, Vue, Svelte, Astro, Tailwind) — also when the user says a page looks generic, templated, \"AI-made\", sloppy, or asks to make it look designed, premium, award-level or less like every other site. Not for dashboards or app screens.","https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Ftaste-design\u002FLICENSE","https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Ftaste-design\u002FSKILL.md",[],[],{"skillNo":85,"name":86,"nameZh":6,"description":87,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":88,"categories":89,"keywords":90,"publishedAt":19},"01M7EWY8FQFNQ2GB7NZZWGSSDX","test-driven-development","Use when implementing any feature or bugfix, before writing implementation code","https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fsuperpowers-tdd\u002FSKILL.md",[],[],{"skillNo":92,"name":93,"nameZh":6,"description":94,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":95,"categories":96,"keywords":97,"publishedAt":19},"01M7EWY5SHBFRH1HDHVWM9K7TV","systematic-debugging","Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes","https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fsuperpowers-debugging\u002FSKILL.md",[],[],{"skillNo":99,"name":100,"nameZh":6,"description":101,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":102,"categories":103,"keywords":104,"publishedAt":19},"01M7EWY3NM0Z0JAKS7HDZ24K66","requesting-code-review","Use when completing tasks, implementing major features, or before merging to verify work meets requirements","https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fsuperpowers-code-review\u002FSKILL.md",[],[],{"skillNo":106,"name":107,"nameZh":6,"description":108,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":109,"categories":110,"keywords":111,"publishedAt":19},"01M7EWY0WRMWX4M9ZFY0WKC593","brainstorming","You MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation).","https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fsuperpowers-brainstorming\u002FSKILL.md",[],[]]