mlx-audio
Use MLX Audio for local TTS, STT, speech enhancement, VAD, realtime speech routing, API serving, and conversion workflows.
- License 许可
- MIT
- 最近更新
- 2026-10-08
- Stars
- ★ 11
- Forks
- 3
01 前言 Front Matter
- name
- mlx-audio
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Use MLX Audio for local TTS, STT, speech enhancement, VAD, realtime speech routing, API serving, and conversion workflows.
- disable-model-invocation
- true
02 资源文件 8
- 参考文档references/api-reference.md2.8 KB
- 参考文档references/audio-io-and-dsp.md1.2 KB
- 参考文档references/cli-reference.md1.7 KB
- 参考文档references/install-and-dependencies.md2.2 KB
- 参考文档references/model-overview.md1.2 KB
- 参考文档references/repo-provenance.md1.2 KB
- 参考文档references/repo-routing-metadata.json494 B
- 参考文档references/troubleshooting.md2.1 KB
03 SKILL.md 正文
MLX Audio Repo Skill
Use this skill for practical work with mlx-audio 0.4.8: package installation, model-family selection, CLI routing, shared audio I/O, speech streaming, server endpoints, and conversion/quantization guidance.
Start Here
- Check the install and core APIs with
scripts/check_install.py. - Read
references/install-and-dependencies.mdfor extras, backend notes, and safe setup. - Use
references/api-reference.mdandreferences/model-overview.mdfor verified package facts. - Route the task to the focused sub-skill below instead of treating this root file as a manual.
Route by Task
- Text-to-speech generation, streaming, voice cloning, or model-specific TTS knobs:
sub-skills/tts-generation/ - Speech-to-text transcription, hotwords, forced alignment, streaming ASR, or WER:
sub-skills/stt-transcription/ - Audio enhancement, source separation, VAD, turn detection, or audio I/O:
sub-skills/speech-transforms-vad/ - OpenAI-compatible server, realtime WebSockets, Studio UI, or conversion/quantization:
sub-skills/server-and-conversion/
Core Safety Defaults
- Prefer package docs and the bundled scripts before launching long jobs.
- Use tiny fixtures, parser checks, and
--helpsmoke tests before downloading weights or starting a server. - Treat real model execution as optional unless the user explicitly wants a full inference run.
- For package import or dependency issues, consult
references/troubleshooting.mdfirst.
Bundled Root References
references/repo-provenance.md: source baseline, package version, dirty state, and evidence pathsreferences/repo-routing-metadata.json: router placement and selection guidancereferences/install-and-dependencies.md: core dependencies, extras, and safe install checksreferences/api-reference.md: verified signatures and shared runtime factsreferences/model-overview.md: supported TTS/STT/STS/VAD familiesreferences/cli-reference.md: CLI names and route mapreferences/audio-io-and-dsp.md: shared audio I/O and DSP behaviorreferences/troubleshooting.md: cross-cutting install, import, dependency, and audio issues
Bundled Root Scripts
scripts/check_install.py: import/version/API smoke checkscripts/check_optional_deps.py: optional dependency and backend probe
Common Entry Points
mlx_audio.tts.generateandmlx_audio.tts.utils.load_modelmlx_audio.stt.generateandmlx_audio.stt.utils.load_modelmlx_audio.sts.generatemlx_audio.servermlx_audio.convert
When to Escalate
If the request needs model weights, device-specific performance claims, or realtime behavior that depends on unavailable hardware, keep the answer scoped to the bundled docs, parser checks, and safe command planning.
04 同仓库更多 Skills
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
python-api
Programmatic Object-Detection-Metrics API guidance for BoundingBox, BoundingBoxes, Evaluator, enums, coordinate conversion, AP/mAP semantics, and source-style import diagnostics.
file-evaluation
Folder-based PASCAL VOC AP/mAP evaluation from ground-truth and detection text files.
object-detection-metrics
Guides legacy Object-Detection-Metrics PASCAL VOC AP/mAP evaluation from detection text folders and direct Python API classes.
nx-workspace-patterns
Configure and optimize Nx monorepo workspaces. Use when setting up Nx, configuring project boundaries, optimizing build caching, or implementing affected commands.