inference.py

import torchaudio
import torch

from transformers import (
    AutoTokenizer,
    AutoModelForCausalLM,
)

from speechtokenizer import SpeechTokenizer
from audiotools import AudioSignal


def decode_tts(tokens, quantizer, n_codebooks, n_original_tokens, start_audio_token_id, end_audio_token_id):
    # find start and end indices of audio tokens
    start = torch.nonzero(tokens == start_audio_token_id)
    end = torch.nonzero(tokens == end_audio_token_id)

    start = start[0, -1] + 1 if len(start) else 0
    end = end[0, -1] if len(end) else tokens.shape[-1]

    # subtract length of original vocabulary -> tokens in range [0, 1024)
    audio_tokens = tokens[start:end] % n_original_tokens
    reminder = audio_tokens.shape[-1] % n_codebooks

    if reminder:
        # pad if last frame is incomplete
        pad_tokens = torch.zeros(n_codebooks - reminder, device="cuda")
        audio_tokens = torch.cat([audio_tokens, pad_tokens[reminder:n_codebooks]], dim=0)

    transposed = audio_tokens.view(-1, n_codebooks).t()
    codes = transposed.view(n_codebooks, 1, -1).to(device)

    audio = quantizer.decode(codes).squeeze(0)

    del tokens
    del audio_tokens
    torch.cuda.empty_cache()

    return AudioSignal(audio.detach().cpu().numpy(), quantizer.sample_rate)


def infer_text_to_audio(text, model, tokenizer, quantizer, max_seq_length=1024, top_k=20):
    text_tokenized = tokenizer(text, return_tensors="pt")
    text_input_tokens = text_tokenized["input_ids"].to(device)

    soa = tokenizer(start_audio_token, return_tensors="pt")["input_ids"][:, -1:].to(device)
    eoa = tokenizer(end_audio_token, return_tensors="pt")["input_ids"][:, -1:].to(device)

    text_tokens = torch.cat([text_input_tokens, soa], dim=1)
    attention_mask = torch.ones(text_tokens.size(), device=device)

    output_audio_tokens = model.generate(
        text_tokens,
        attention_mask=attention_mask,
        max_new_tokens=max_seq_length,
        top_k=top_k,
        do_sample=True,
        temperature=0.8,
        no_repeat_ngram_size=3,
    )

    audio_signal = decode_tts(output_audio_tokens[0], quantizer, 3, len(tokenizer) - codebook_size, soa, eoa)

    return audio_signal


def infer_audio_to_text(audio_path, model, tokenizer, quantizer, max_seq_length=1024, top_k=20):
    audio_data, sample_rate = torchaudio.load(audio_path)

    audio = audio_data.view(1, 1, -1).float().to(device)
    # bandwidth_id = torch.tensor([0])

    codes = quantizer.encode(audio)
    raw_audio_tokens = codes[:, :n_codebooks_asr] + len(tokenizer) - codebook_size

    soa = tokenizer(start_audio_token, return_tensors="pt")["input_ids"][:, -1:].to(device)
    eoa = tokenizer(end_audio_token, return_tensors="pt")["input_ids"][:, -1:].to(device)
    audio_tokens = torch.cat([soa, raw_audio_tokens.view(1, -1), eoa], dim=1)
    tokens = torch.cat([audio_tokens], dim=1)

    attention_mask = torch.ones(tokens.size(), device=device)

    output_text_tokens = model.generate(
        tokens,
        attention_mask=attention_mask,
        max_new_tokens=max_seq_length,
        temperature=0.6,
        top_p=0.9,
        top_k=top_k,
        no_repeat_ngram_size=4,
        length_penalty=2.0,
        repetition_penalty=1.5
    )

    output_text_tokens = output_text_tokens.cpu()[0]
    output_text_tokens = output_text_tokens[output_text_tokens < tokenizer(start_audio_token)["input_ids"][-1]]
    decoded_text = tokenizer.decode(output_text_tokens, skip_special_tokens=True)

    return decoded_text


device = "cuda"

n_special_tokens = 3
n_codebooks_tts = 3
n_codebooks_asr = 1

start_audio_token = "<soa>"
end_audio_token = "<eoa>"
end_sequence_token = "<eos>"

base_model = "Vikhrmodels/salt-116k"


if __name__ == "__main__":
    tokenizer = AutoTokenizer.from_pretrained(base_model, cache_dir=".")
    model = AutoModelForCausalLM.from_pretrained(
        base_model,
        cache_dir=".",
        torch_dtype=torch.bfloat16,
        attn_implementation="sdpa",
        device_map={"": 0}
    )

    quantizer_speech = SpeechTokenizer.load_from_checkpoint("speechtokenizer/config.json",
                                                            "speechtokenizer/SpeechTokenizer.pt")
    quantizer_speech = quantizer_speech.eval().to(device)
    codebook_size = quantizer_speech.quantizer.bins

    text = ("Say 'COUNT NUMBERS FROM ONE TO TEN' with a male speaker delivers a very monotone and "
            "low-pitched speech with a moderate speed in a setting with almost no noise, "
            "creating a clear and quiet recording.")

    audio_signal = infer_text_to_audio(text, model, tokenizer, quantizer_speech, top_k=50)
    audio_signal.write("output.wav")

    audio_path = "./input.wav"
    generated_text = infer_audio_to_text(audio_path, model, tokenizer, quantizer_speech)
    print(generated_text)