Listen Lab
The Listen Lab is a full translation and re-read pipeline. Drop in an audio file, transcribe it, translate it, and hear it spoken back in your target language — all locally, with your chosen voice.
The Listen Lab chains three steps into one workflow:
Audio File
│
▼
Whisper — transcribes in the source language
│
▼
Local LLM — translates to your target language
│
▼
TTS Engine — speaks the translated text with your chosen voice
│
▼
Audio Output (playable and saveable)
This is useful for:
- Listening to foreign-language podcasts or interviews in your own language
- Re-narrating audio content in a different language using a cloned voice
- Creating translated audiobook versions from existing recordings
- Drag and drop audio files onto the playlist (
.wav,.mp3,.m4a,.flac,.ogg) - Enable Translate & Re-read toggle
- Select your target language
- Select a voice for the TTS output
- Click Play Selected or Play All
The pipeline runs automatically. Playback begins streaming as soon as the first TTS chunks are generated — you don't wait for the full file to be processed.
| Control | Description |
|---|---|
| Translate & Re-read | Toggle the translation pipeline on/off |
| Target Language | Language to translate into |
| Voice | Which voice the TTS engine uses for the re-read |
| Volume | Output volume (shared with Speech Lab) |
| Pause / Resume | Pauses during the TTS phase |
| Save | Saves the translated audio to a file |
| Play Selected | Processes and plays checked items in order |
Each item in the playlist shows its current status:
| Status | Meaning |
|---|---|
| Queued | Waiting to process |
| Transcribing | Whisper is running |
| Translating | LLM is translating |
| Converting | TTS is generating audio |
| Done | Complete — audio available |
| Error | Something went wrong — check the log |
After processing, click Save on any completed item to export the translated audio as MP3 or WAV.
- The Listen Lab uses the same LLM as the rest of the app for translation. For long audio files, a larger model (Gemma 3 1B or 4B) produces more natural-sounding translations — see LLM Models.
- If you want the re-read to use a cloned voice, make sure you have voice profiles set up in the Voice Lab and Fish-Speech is your active engine.
- For Kokoro users, select a preset voice that matches your target language for the most natural result (e.g. a Japanese preset voice for Japanese output).
- The transcription step uses the Whisper model configured in Settings. If your source audio is in a language Whisper handles poorly at "base" size, try upgrading to "small" or "medium" in Settings before running the pipeline.