Your first Chinese transcript¶
Test a short, clear Chinese recording before processing long episodes or a subscription backlog. Complete installation and models/tools first.
1. Check readiness¶
atot doctor
atot models
Resolve required CLI, backend, model, and FFmpeg/ffprobe failures. Optional Cloud Summary can remain unconfigured for local ASR and MCP.
Use Settings → Overview in the native app. Qt offers Doctor and an external CLI path. See troubleshooting if app and terminal disagree.
2. Import one source¶
Choose New Item (Command-N) in the native app and select local media or a supported episode URL and group. Import establishes the source; processing creates text. Subscription metadata alone does not mean audio or a transcript exists.
atot exec transcribe path/to/episode.m4a --language zh --local-only --json
atot exec import "https://www.xiaoyuzhoufm.com/episode/<episode-id>" \
--language zh --local-only --json
Replace paths and IDs. URL imports contact the source for metadata/media; ASR then runs locally. Apple Podcasts show and RSS URLs use subscriptions; exec import takes a Xiaoyuzhou episode URL.
3. Select the Chinese hint¶
In current source, select Chinese / Mandarin under Settings → Local ASR → Transcription Languages to reproduce the validated workflow.
No selection means auto; multiple selections use auto plus mixed-language hints and retain original speech. New configurations default to auto, without claiming equal validation for other languages. Explicit old choices remain.
v0.9.0 lacks the new checkboxes and output selector but accepts a single --language zh. See languages.
4. Run locally¶
Select Transcribe locally only, or Command-Return. Follow the stage and log: download completion, one ASR chunk, and final success are different states. Do not repeatedly start the same long task.
The CLI examples perform transcription and QA with --local-only, skipping optional cloud summaries and requiring no model key.
5. Read and check¶
Open Transcript; Overview preserves metadata. Summary has content after a summary workflow runs. Use the actual returned work directory:
atot exec qa path/to/work-dir --json
atot exec repair path/to/work-dir --plan-only --json
atot exec library status --json
| Output | Purpose |
|---|---|
| transcript.readable.md | Reading text |
| transcript.timestamped.md | Locate source passages |
| transcript.srt / transcript.vtt | Captions |
| qa.json | Automated findings |
| corrections.json | Change audit |
| Library documents/transcript.md | Published reading document |
Check important names, numbers, judgments, and quotations against audio. QA is not an accuracy guarantee. Inspect a repair plan before rerunning windows.
6. Read a summary when useful¶
Choose Codex/MCP, using the current conversation model without a SonaFolio key, or cloud API summaries. Have the agent read the full source, retain timestamps, and disclose uncertainty. You can start your own reading with its summary.
After the first run, try subscriptions, subtitles, or iOS reading.