AI GitHub
F5-TTS for reference-conditioned synthesis
Generate new speech using reference audio and text through the published model pipeline.
Key Takeaways
- The system conditions synthesis on a reference sample and text.
- Prepare the reference audio and accurate matching text expected by the selected workflow.
- Use a clean, authorized reference with an accurate transcript and generate a short sentence.
- Reference transcription errors can affect generation.
How it works
The system conditions synthesis on a reference sample and text. Official examples show how the model and vocoder fit together for inference.
What you need
Prepare the reference audio and accurate matching text expected by the selected workflow. Use a compatible model and vocoder before experimenting with custom voices.
What to try first
Use a clean, authorized reference with an accurate transcript and generate a short sentence. Repeat with a deliberately noisy reference and compare errors.
Check reference audio with a long pause and output text with a number. Transcription alignment and text normalization can each influence how the line is read.
Where to look for failures
Reference transcription errors can affect generation. Confirm artifact licenses and inspect output for dropped or invented words.
Related Franklin pages
Use authorized voice references and compare both intelligibility and similarity. Check the exact checkpoint terms before moving from a local test to published audio.
- Try supported multilingual voice synthesis and reference-conditioned modes. CosyVoice for multilingual speech generation
- Generate speech and experiment with reference audio and documented expression controls. Chatterbox for controllable text-to-speech
For an application where the quality of recording and handling audio matters, see Franklin’s Open Mic AI profile. Open Mic AI
Project documentation
Try it for your workflow
Explore the project’s documentation for setup and usage, or save this page to a collection for later.
Read the project documentation ↗