What problem does it solve?
Choosing between Step ASR and Volcengine Doubao ASR for local audio/video transcription is error-prone: picking the wrong backend wastes paid quota, loses timestamp or SRT capability, and risks double billing when a task is resubmitted to another provider. This Skill acts as a unified routing layer that selects the correct backend, runs a read-only readiness check, and enforces billing-safety rules.
Core Features & Use Cases
- Offline backend routing: The route_asr.py script picks Step ASR for plain-text transcription and Doubao ASR for timestamps, SRT subtitles, or talking-head rough cuts, without calling any paid API.
- Read-only readiness check: asr_doctor.py reports whether ffmpeg, scripts, App IDs, and tokens are present without printing secrets or calling paid endpoints.
- Billing and fallback safety: Once a request is submitted to one provider, the Skill forbids cross-provider resubmission and requires resuming via request ID, cache, or --resume.
- Use Case: You have a local MP4 interview and need SRT subtitles. The router selects Doubao ASR, the doctor confirms credentials and ffmpeg, and the existing yichen-volc-asr executor produces the transcript and subtitle files.
Quick Start
Ask the assistant to transcribe a local audio or video file with $yichen-asr, stating whether you need plain text, timestamps, SRT subtitles, or a rough cut.