---

name: stt-transcription description: "Use MLX Audio STT workflows for transcription, streaming ASR, hotwords, forced alignment, and WER evaluation." disable-model-invocation: true metadata: disco-role: operating license: MIT

STT Transcription

Use this sub-skill when the user wants to transcribe audio, stream transcription results, inject hotwords or context, run forced alignment, or compute WER summaries.

Route Here For

• CLI or Python speech-to-text generation.
• Streaming transcription and partial results.
• Context or hotword guidance.
• Forced alignment with an explicit transcript.
• WER evaluation and summary planning.
• Safe planning for mlx_audio.stt.generate and mlx_audio.stt.eval.

Route Elsewhere

• For text-to-speech generation or voice cloning, use ../tts-generation/.
• For audio enhancement, separation, VAD, or audio I/O, use ../speech-transforms-vad/.
• For server endpoints or conversion, use ../server-and-conversion/.

Fast Paths

• See references/workflows.md for transcription, alignment, streaming, and WER recipes.
• See references/api-reference.md for stable kwargs and output-format facts.
• See references/troubleshooting.md for audio path, JSON, and alignment failures.
• Use scripts/stt_command_builder.py to shape a safe CLI command.
• Use scripts/wer_summary.py to summarize simple WER pairs or line-aligned fixtures.

Default Safety Policy

Confirm the audio path, output format, and whether the model needs a transcript or context before a long run. If the user only needs a command plan, prefer the bundled builders and docs over a live inference call.