app.py

import spaces
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
import torch
import soundfile as sf
from xcodec2.modeling_xcodec2 import XCodec2Model
import torchaudio
import gradio as gr
import tempfile

llasa_3b ='srinivasbilla/llasa-3b'

tokenizer = AutoTokenizer.from_pretrained(llasa_3b)

model = AutoModelForCausalLM.from_pretrained(
    llasa_3b,
    trust_remote_code=True,
    use_cache=False,
    torch_dtype=torch.bfloat16,
    device_map='cuda',
    return_dict=True
)

model_path = "srinivasbilla/xcodec2"
 
Codec_model = XCodec2Model.from_pretrained(model_path)
Codec_model.eval().cuda()

whisper_turbo_pipe = pipeline(
    "automatic-speech-recognition",
    model="openai/whisper-large-v3-turbo",
    torch_dtype=torch.float16,
    device='cuda',
)

def ids_to_speech_tokens(speech_ids):
 
    speech_tokens_str = []
    for speech_id in speech_ids:
        speech_tokens_str.append(f"<|s_{speech_id}|>")
    return speech_tokens_str

def extract_speech_ids(speech_tokens_str):
 
    speech_ids = []
    for token_str in speech_tokens_str:
        if token_str.startswith('<|s_') and token_str.endswith('|>'):
            num_str = token_str[4:-2]

            num = int(num_str)
            speech_ids.append(num)
        else:
            print(f"Unexpected token: {token_str}")
    return speech_ids

@spaces.GPU(duration=60)
def infer(sample_audio_path, target_text, progress=gr.Progress()):
    with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as f:
        progress(0, 'Loading and trimming audio...')
        waveform, sample_rate = torchaudio.load(sample_audio_path)
        if len(waveform[0])/sample_rate > 15:
            gr.Warning("Trimming audio to first 15secs.")
            waveform = waveform[:, :sample_rate*15]

        # Check if the audio is stereo (i.e., has more than one channel)
        if waveform.size(0) > 1:
            # Convert stereo to mono by averaging the channels
            waveform_mono = torch.mean(waveform, dim=0, keepdim=True)
        else:
            # If already mono, just use the original waveform
            waveform_mono = waveform

        prompt_wav = torchaudio.transforms.Resample(orig_freq=sample_rate, new_freq=16000)(waveform_mono)
        prompt_text = whisper_turbo_pipe(prompt_wav[0].numpy())['text'].strip()
        progress(0.5, 'Transcribed! Generating speech...')
            
        input_text = prompt_text + ' ' + target_text

        #TTS start!
        with torch.no_grad():
            # Encode the prompt wav
            vq_code_prompt = Codec_model.encode_code(input_waveform=prompt_wav)

            vq_code_prompt = vq_code_prompt[0,0,:]
            # Convert int 12345 to token <|s_12345|>
            speech_ids_prefix = ids_to_speech_tokens(vq_code_prompt)

            formatted_text = f"<|TEXT_UNDERSTANDING_START|>{input_text}<|TEXT_UNDERSTANDING_END|>"

            # Tokenize the text and the speech prefix
            chat = [
                {"role": "user", "content": "Convert the text to speech:" + formatted_text},
                {"role": "assistant", "content": "<|SPEECH_GENERATION_START|>" + ''.join(speech_ids_prefix)}
            ]

            input_ids = tokenizer.apply_chat_template(
                chat, 
                tokenize=True, 
                return_tensors='pt', 
                continue_final_message=True
            )
            input_ids = input_ids.to('cuda')
            speech_end_id = tokenizer.convert_tokens_to_ids('<|SPEECH_GENERATION_END|>')

            # Generate the speech autoregressively
            outputs = model.generate(
                input_ids,
                max_length=2500,  # We trained our model with a max length of 2048
                eos_token_id= speech_end_id ,
                do_sample=True,
                top_p=1,           
                temperature=0.8
            )
            # Extract the speech tokens
            generated_ids = outputs[0][input_ids.shape[1]-len(speech_ids_prefix):-1]

            speech_tokens = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)   

            # Convert  token <|s_23456|> to int 23456 
            speech_tokens = extract_speech_ids(speech_tokens)

            speech_tokens = torch.tensor(speech_tokens).cuda().unsqueeze(0).unsqueeze(0)

            # Decode the speech tokens to speech waveform
            gen_wav = Codec_model.decode_code(speech_tokens) 

            # if only need the generated part
            gen_wav = gen_wav[:,:,prompt_wav.shape[1]:]

            progress(1, 'Synthesized!')

        return (16000, gen_wav[0, 0, :].cpu().numpy())

with gr.Blocks() as app_tts:
    gr.Markdown("# Zero Shot Voice Clone TTS")
    ref_audio_input = gr.Audio(label="Reference Audio", type="filepath")
    gen_text_input = gr.Textbox(label="Text to Generate", lines=10)

    generate_btn = gr.Button("Synthesize", variant="primary")

    audio_output = gr.Audio(label="Synthesized Audio")

    generate_btn.click(
        infer,
        inputs=[
            ref_audio_input,
            gen_text_input,
        ],
        outputs=[audio_output],
    )

with gr.Blocks() as app_credits:
    gr.Markdown("""
# Credits

* [zhenye234](https://github.com/zhenye234) for the original [repo](https://github.com/zhenye234/LLaSA_training)
* [mrfakename](https://huggingface.co/mrfakename) for the [gradio demo code](https://huggingface.co/spaces/mrfakename/E2-F5-TTS)        
""")

with gr.Blocks() as app:
    gr.Markdown(
        """
# llasa 3b TTS

This is a local web UI for llasa 3b SOTA(imo) Zero Shot Voice Cloning and TTS model.

The checkpoints support English and Chinese.

If you're having issues, try converting your reference audio to WAV or MP3, clipping it to 15s, and shortening your prompt.
"""
    )
    gr.TabbedInterface([app_tts], ["TTS"])


app.launch()