| # Core dependencies for tiny-audio model inference | |
| # This file is pushed to HuggingFace for model repository | |
| # Transcription and word timestamps (Qwen3-ForcedAligner) need transformers >= 5.17. | |
| # Speaker diarization (return_speakers=True) uses Nemotron-3-Diarization, which is in | |
| # transformers main after 5.17 -- pinned to the commit the pipeline was verified on. | |
| transformers @ git+https://github.com/huggingface/transformers@91ea1192a00e0ebb684400434375fbd2345c8449 | |
| torch | |
| torchaudio | |
| peft | |
| # Nemotron's streaming feature extractor (return_speakers=True) imports it. | |
| librosa | |