whisper-large-russian

Usage

In order to process phone calls it is highly recommended that you preprocess your records and adjust volume before performing ASR. For example, like this:

sox record.wav -r 16k record-normalized.wav norm -0.5 compand 0.3,1 -90,-90,-70,-70,-60,-20,0,0 -5 0 0.2

Then your ASR code should look somewhat like this:

import torch
from transformers import WhisperForConditionalGeneration, WhisperProcessor, pipeline

torch_dtype = torch.bfloat16 # set your preferred type here

device = 'cpu'
if torch.cuda.is_available():
    device = 'cuda'
elif torch.backends.mps.is_available():
    device = 'mps'
    setattr(torch.distributed, "is_initialized", lambda : False) # monkey patching
device = torch.device(device)

whisper = WhisperForConditionalGeneration.from_pretrained(
    "antony66/whisper-large-v3-russian", torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True,
    # add attn_implementation="flash_attention_2" if your GPU supports it
)

processor = WhisperProcessor.from_pretrained("antony66/whisper-large-v3-russian")

asr_pipeline = pipeline(
    "automatic-speech-recognition",
    model=whisper,
    tokenizer=processor.tokenizer,
    feature_extractor=processor.feature_extractor,
    max_new_tokens=256,
    chunk_length_s=30,
    batch_size=16,
    return_timestamps=True,
    torch_dtype=torch_dtype,
    device=device,
)

# read your wav file into variable wav. For example:
from io import BufferIO
wav = BytesIO()
with open('record-normalized.wav', 'rb') as f:
    wav.write(f.read())
wav.seek(0)

# get the transcription
asr = asr_pipeline(wav, generate_kwargs={"language": "russian", "max_new_tokens": 256}, return_timestamps=False)

print(asr['text'])

Work in progress

This model is in WIP state for now. The goal is to finetune it for speech recognition of phone calls as much as possible. If you want to contribute and you know or have any good dataset please let me know. Your help will be much appreciated.