Fabio
c36510c21e
Some checks failed
Linux Build / Linux cpu (push) Has been cancelled
Linux Build / Linux cpu-clang (push) Has been cancelled
Linux Build / Linux vulkan (push) Has been cancelled
mac-build / macOS CPU (push) Has been cancelled
Nix Build / Nix (cpu) (push) Has been cancelled
Nix Build / Nix (vulkan) (push) Has been cancelled
windows-build / Windows CPU (push) Has been cancelled
18 lines
983 B
Markdown
18 lines
983 B
Markdown
# Parakeet TDT test fixtures
|
|
|
|
- `2086-149220-0033.wav`
|
|
- 16 kHz mono, ~7.44 s speech clip, extracted from the LibriSpeech
|
|
`test-clean` corpus, utterance `2086-149220-0033`
|
|
("Well, I don't wish to see it any more, observed Phoebe, turning away
|
|
her eyes. It is certainly very like the old portrait.").
|
|
- LibriSpeech (Panayotov et al., 2015) is distributed under CC BY 4.0:
|
|
https://www.openslr.org/12
|
|
- Used by `test_golden_transcription.cpp` and
|
|
`test_streaming_transcription.cpp` as a fixed-input regression check for
|
|
offline full-context, bounded-window long-form, word timestamps, buffered
|
|
streaming, finalize, and reset. Both tests accept the installed
|
|
safetensors directory or a standalone GGUF through `--model`.
|
|
The expected transcription was verified against the actual NeMo
|
|
reference model (`nvidia/parakeet-tdt-0.6b-v3` via
|
|
`nemo.collections.asr.models.ASRModel.from_pretrained`) for this exact
|
|
clip, not just eyeballed.
|