asr
Use for NeMo Speech ASR model selection, checkpoint loading, transcription, timestamps, streaming, fine-tuning, Lhotse batching, decoding customization, and ASR evaluation.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- asr
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Use for NeMo Speech ASR model selection, checkpoint loading, transcription, timestamps, streaming, fine-tuning, Lhotse batching, decoding customization, and ASR evaluation.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/api-reference.md10.1 KB
- 参考文档references/configuration.md12.3 KB
- 参考文档references/troubleshooting.md11.8 KB
- 参考文档references/workflows.md14.0 KB
- 脚本scripts/check_asr_manifest.py13.0 KB
03 SKILL.md 正文
ASR
Use this sub-skill for NeMo Speech automatic speech recognition tasks: selecting CTC/RNNT/TDT/Hybrid/AED Canary models, loading .nemo checkpoints or pretrained names, offline transcription, timestamps, long-audio inference, cache-aware or buffered streaming, fine-tuning, Lhotse/dynamic bucketing, tokenizer changes, decoding customization, confidence/LM/word boosting, and WER/BLEU-style ASR evaluation.
Route elsewhere when the task is primarily speaker diarization, VAD, or forced alignment; generic manifest sharding/tokenizer utilities; repository testing/formatting; or SpeechLM2/SALM duplex work. For generic data tooling, prefer ../data-tools/SKILL.md when present; for repo workflows, prefer ../repo-development/SKILL.md when present; for SpeechLM2, prefer ../speechlm2/SKILL.md when present.
Start Here
- Read
references/workflows.mdfor task-specific command and Python patterns. - Read
references/api-reference.mdbefore using model classes,transcribe(), timestamps, streaming pipeline builders, or decoding APIs. - Read
references/configuration.mdbefore editing Hydra YAML, manifests, Lhotse settings, tokenizer changes, Canary prompts, LM/customization, or evaluation settings. - Read
references/troubleshooting.mdwhen installs, CUDA/backends, Hydra overrides, manifests, Lhotse, streaming, decoding, or fine-tuning fail.
Safe Bundled Tools
scripts/check_asr_manifest.pyvalidates small or large ASR JSONL manifests without importing NeMo, downloading models, touching audio contents, or writing outputs unless explicitly redirected by the caller.- Run it before fine-tuning or evaluation manifests:
python scripts/check_asr_manifest.py train.jsonl --canary --min-duration 0.1 --max-duration 30.
Evidence Base
This sub-skill distills repository evidence from docs/source/asr/inference.rst, docs/source/asr/fine_tuning.rst, docs/source/asr/configs.rst, docs/source/asr/datasets.rst, docs/source/asr/asr_language_modeling_and_customization.rst, docs/source/dataloaders.rst, ASR example scripts/configs, ASR model/inference source, ASR tests, and the prior .claude/skills/nemo-speech-asr-finetune/SKILL.md. Runtime guidance is self-contained; do not require future agents to reopen those source files.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).