Can't Transcribe Telugu audio files

[1preethi]

Checks

• This template is only for bug reports, usage problems go with 'Help Wanted'.
• I have thoroughly reviewed the project documentation but couldn't find information to solve my problem.
• I have searched for existing issues, including closed ones, and couldn't find a solution.
• I am using English to submit this issue to facilitate community communication.

Environment Details

Python 3.10, cuda 12.8, Ubuntu 22.04

Steps to Reproduce

1. pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128
2. clone and navigate to the repository
3. pip install -e .
4. f5-tts_finetune-gradio --share

✔️ Expected Behavior

Transcripts should be generated successfully

❌ Actual Behavior

Using chunk_length_s is very experimental with seq2seq models. The results will not necessarily be entirely accurate and will have caveats. More information: huggingface/transformers#20104. Ignore this warning with pipeline(..., ignore_warning=True). To use Whisper for long-form transcription, use rather the model's generate method directly as the model relies on it's own chunking mechanism (cf. Whisper original paper, section 3.8. Long-form Transcription).

Error:

transcribe complete samples : 0
path : /teamspace/studios/this_studio/F5-TTS/src/f5_tts/../../data/test_telugu_pinyin/wavs
error files : 1 (only 1 file is in dataset)

[onestardao]

You’ve hit a classic failure mode we’ve seen over and over again — your issue maps directly to Problem No.3 on the AI Failure Map:

Chunking failure due to incompatible segmentation strategy.

In short, you're feeding audio into a system that expects a totally different chunking regime. Whisper-style models (and other seq2seq systems) assume their own built-in chunking logic — and when you force it with chunk_length_s, the semantic context gets shredded.

If you're trying to fix this at the root, I’ve documented a full solution with architecture-level fixes and real-world working code in my open-source map:
🔍 WFGY ProblemMap – No.3: Chunking Breakage

I’ve seen this exact failure cause hallucinations, null outputs, or complete crashes across audio + PDF + text modalities. Let me know if you want the exact patch pattern for long-form audio. I’ve got it. You want it?

[1preethi]

Yes @onestardao . Could you please help me resolve it?

[onestardao]

Sure — you can start by grabbing the TXT OS package
It comes with the symbolic math overlay (WFGY Engine) + all the formulas and quick prompts.

Once downloaded, just tell your AI:

“Use the WFGY engine in TXTOS to fix Problem No.3: chunking failure across audio input.
Here’s my setup: [describe your audio format + model + pipeline briefly]. What patch pattern do I need?”

We’ve already helped 70+ devs solve this exact chunking issue, especially across Whisper + seq2seq conflicts.
Let me know how it goes — happy to help troubleshoot if anything’s unclear.