snoopycn / whisper-plus

WhisperPlus: Advancing Speech-to-Text Processing 🚀

WhisperPlus: Advancing Speech2Text and Text2Speech Feature 🚀

🛠️ Installation

pip install whisperplus git+https://github.com/huggingface/transformers
pip install flash-attn --no-build-isolation

🤗 Model Hub

You can find the models on the HuggingFace Model Hub

🎙️ Usage

To use the whisperplus library, follow the steps below for different tasks:

🎵 Youtube URL to Audio

from whisperplus import SpeechToTextPipeline, download_and_convert_to_mp3
from transformers import BitsAndBytesConfig, HqqConfig


url = "https://www.youtube.com/watch?v=di3rHkEZuUw"
audio_path = download_and_convert_to_mp3(url)

hqq_config = HqqConfig(
    nbits=1,
    group_size=64,
    quant_zero=False,
    quant_scale=False,
    axis=0,
    offload_meta=False,
)  # axis=0 is used by default

bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16,
    bnb_4bit_use_double_quant=True,
)

pipeline = SpeechToTextPipeline(
    model_id="distil-whisper/distil-large-v3", quant_config=hqq_config
)  # or bnb_config

transcript = pipeline(
    audio_path=audio_path,
    chunk_length_s=30,
    stride_length_s=5,
    max_new_tokens=128,
    batch_size=100,
    language="english",
    return_timestamps=False,
)

print(transcript)

📰 Summarization

from whisperplus import TextSummarizationPipeline

summarizer = TextSummarizationPipeline(model_id="facebook/bart-large-cnn")
summary = summarizer.summarize(transcript)
print(summary[0]["summary_text"])

📰 Long Text Support Summarization

from whisperplus import LongTextSummarizationPipeline

summarizer = LongTextSummarizationPipeline(model_id="facebook/bart-large-cnn")
summary_text = summarizer.summarize(transcript)
print(summary_text)

💬 Speaker Diarization

from whisperplus import (
    ASRDiarizationPipeline,
    download_and_convert_to_mp3,
    format_speech_to_dialogue,
)

audio_path = download_and_convert_to_mp3("https://www.youtube.com/watch?v=mRB14sFHw2E")

device = "cuda"  # cpu or mps
pipeline = ASRDiarizationPipeline.from_pretrained(
    asr_model="openai/whisper-large-v3",
    diarizer_model="pyannote/speaker-diarization",
    use_auth_token=False,
    chunk_length_s=30,
    device=device,
)

output_text = pipeline(audio_path, num_speakers=2, min_speaker=1, max_speaker=2)
dialogue = format_speech_to_dialogue(output_text)
print(dialogue)

⭐ RAG - Chat with Video(LanceDB)

from whisperplus.pipelines.chatbot import ChatWithVideo

chat = ChatWithVideo(
    input_file="trascript.txt",
    llm_model_name="TheBloke/Mistral-7B-v0.1-GGUF",
    llm_model_file="mistral-7b-v0.1.Q4_K_M.gguf",
    llm_model_type="mistral",
    embedding_model_name="sentence-transformers/all-MiniLM-L6-v2",
)

query = "what is this video about ?"
response = chat.run_query(query)
print(response)

🌠 RAG - Chat with Video(AutoLLM)

from whisperplus import AutoLLMChatWithVideo

# service_context_params
system_prompt = """
You are an friendly ai assistant that help users find the most relevant and accurate answers
to their questions based on the documents you have access to.
When answering the questions, mostly rely on the info in documents.
"""
query_wrapper_prompt = """
The document information is below.
---------------------
{context_str}
---------------------
Using the document information and mostly relying on it,
answer the query.
Query: {query_str}
Answer:
"""

chat = AutoLLMChatWithVideo(
    input_file="input_dir",  # path of mp3 file
    openai_key="YOUR_OPENAI_KEY",  # optional
    huggingface_key="YOUR_HUGGINGFACE_KEY",  # optional
    llm_model="gpt-3.5-turbo",
    llm_max_tokens="256",
    llm_temperature="0.1",
    system_prompt=system_prompt,
    query_wrapper_prompt=query_wrapper_prompt,
    embed_model="huggingface/BAAI/bge-large-zh",  # "text-embedding-ada-002"
)

query = "what is this video about ?"
response = chat.run_query(query)
print(response)

🎙️ Speech to Text

from whisperplus import TextToSpeechPipeline

tts = TextToSpeechPipeline(model_id="suno/bark")
audio = tts(text="Hello World", voice_preset="v2/en_speaker_6")

🎥 AutoCaption

from whisperplus import WhisperAutoCaptionPipeline

caption = WhisperAutoCaptionPipeline(model_id="openai/whisper-large-v3")
caption(video_path="test.mp4", output_path="output.mp4", language="turkish")

😍 Contributing

pip install -r dev-requirements.txt
pre-commit install
pre-commit run --all-files

📜 License

This project is licensed under the terms of the Apache License 2.0.

🤗 Citation

@misc{radford2022whisper,
  doi = {10.48550/ARXIV.2212.04356},
  url = {https://arxiv.org/abs/2212.04356},
  author = {Radford, Alec and Kim, Jong Wook and Xu, Tao and Brockman, Greg and McLeavey, Christine and Sutskever, Ilya},
  title = {Robust Speech Recognition via Large-Scale Weak Supervision},
  publisher = {arXiv},
  year = {2022},
  copyright = {arXiv.org perpetual, non-exclusive license}
}