app2.py

print("WARNING: You are running this unofficial E2/F5 TTS demo locally, it may not be as up-to-date as the hosted version (https://huggingface.co./spaces/mrfakename/E2-F5-TTS)")

import os
import re
import torch
import torchaudio
import gradio as gr
import numpy as np
import tempfile
from einops import rearrange
from ema_pytorch import EMA
from vocos import Vocos
from pydub import AudioSegment
from model import CFM, UNetT, DiT, MMDiT
from cached_path import cached_path
from model.utils import (
    get_tokenizer,
    convert_char_to_pinyin,
    save_spectrogram,
)
from transformers import pipeline
import librosa
import re
import gc
import matplotlib.pyplot as plt
import devicetorch

device = "cuda" if torch.cuda.is_available() else "mps" if torch.backends.mps.is_available() else "cpu"

gc.collect()
devicetorch.empty_cache(torch)

print(f"Using {device} device")


# --------------------- Settings -------------------- #

target_sample_rate = 24000
n_mel_channels = 100
hop_length = 256
target_rms = 0.1
nfe_step = 32  # 16, 32
cfg_strength = 2.0
ode_method = 'euler'
sway_sampling_coef = -1.0
speed = 1.0
# fix_duration = 27  # None or float (duration in seconds)
fix_duration = None

def load_model(repo_name, exp_name, model_cls, model_cfg, ckpt_step):
    checkpoint = torch.load(str(cached_path(f"hf://SWivid/{repo_name}/{exp_name}/model_{ckpt_step}.pt")), map_location=device)

    vocab_char_map, vocab_size = get_tokenizer("Emilia_ZH_EN", "pinyin")
    model = CFM(
        transformer=model_cls(
            **model_cfg,
            text_num_embeds=vocab_size,
            mel_dim=n_mel_channels
        ),
        mel_spec_kwargs=dict(
            target_sample_rate=target_sample_rate,
            n_mel_channels=n_mel_channels,
            hop_length=hop_length,
        ),
        odeint_kwargs=dict(
            method=ode_method,
        ),
        vocab_char_map=vocab_char_map,
    ).to(device)

    ema_model = EMA(model, include_online_model=False).to(device)
    ema_model.load_state_dict(checkpoint['ema_model_state_dict'])
    ema_model.copy_params_from_ema_to_model()

    return model

# load models
F5TTS_model_cfg = dict(dim=1024, depth=22, heads=16, ff_mult=2, text_dim=512, conv_layers=4)
E2TTS_model_cfg = dict(dim=1024, depth=24, heads=16, ff_mult=4)

F5TTS_ema_model = load_model("F5-TTS", "F5TTS_Base", DiT, F5TTS_model_cfg, 1200000)
E2TTS_ema_model = load_model("E2-TTS", "E2TTS_Base", UNetT, E2TTS_model_cfg, 1200000)


def chunk_text(text, max_chars=200):
    chunks = []
    current_chunk = ""
    sentences = re.split(r'(?<=[.!?])\s+', text)

    for sentence in sentences:
        if len(current_chunk) + len(sentence) <= max_chars:
            current_chunk += sentence + " "
        else:
            if current_chunk:
                chunks.append(current_chunk.strip())
            current_chunk = sentence + " "

    if current_chunk:
        chunks.append(current_chunk.strip())

    return chunks

def save_spectrogram(y, sr, path):
    plt.figure(figsize=(10, 4))
    D = librosa.amplitude_to_db(np.abs(librosa.stft(y)), ref=np.max)
    librosa.display.specshow(D, sr=sr, x_axis='time', y_axis='hz')
    plt.colorbar(format='%+2.0f dB')
    plt.title('Spectrogram')
    plt.tight_layout()
    plt.savefig(path)
    plt.close()

def infer(ref_audio_orig, ref_text, gen_text, exp_name, remove_silence):
    print(gen_text)
    chunks = chunk_text(gen_text)

    if not chunks:
        raise gr.Error("Please enter some text to generate.")

    # Convert reference audio
    gr.Info("Converting reference audio...")
    with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as f:
        aseg = AudioSegment.from_file(ref_audio_orig)
        aseg = aseg.set_channels(1)
        audio_duration = len(aseg)
        if audio_duration > 15000:
            gr.Warning("Audio is over 15s, clipping to only first 15s.")
            aseg = aseg[:15000]
        aseg.export(f.name, format="wav")
        ref_audio = f.name

    # Select model
    if exp_name == "F5-TTS":
        ema_model = F5TTS_ema_model
#        base_model = F5TTS_base_model
    elif exp_name == "E2-TTS":
        ema_model = E2TTS_ema_model
#        base_model = E2TTS_base_model

    # Transcribe reference audio if needed
    if not ref_text.strip():
        gr.Info("No reference text provided, transcribing reference audio...")
        # Initialize Whisper model
        pipe = pipeline(
            "automatic-speech-recognition",
            model="openai/whisper-large-v3-Turbo", # You can set this to large-V3 if you want better quality, but VRAM then goes to 10 GB
            torch_dtype=torch.float16,
            device=device,
        )
        ref_text = pipe(
            ref_audio,
            chunk_length_s=30,
            batch_size=128,
            generate_kwargs={"task": "transcribe"},
            return_timestamps=False,
        )['text'].strip()
        print("\nTranscribed text: ", ref_text) # Degug transcribing quality
        gr.Info("\nFinished transcription")
        # Release Whisper model
        del pipe
        devicetorch.empty_cache(torch)
        gc.collect()
    else:
        gr.Info("Using custom reference text...")

    # Load and preprocess reference audio
    audio, sr = torchaudio.load(ref_audio)
    if audio.shape[0] > 1:
        audio = torch.mean(audio, dim=0, keepdim=True) # convert to mono
    rms = torch.sqrt(torch.mean(torch.square(audio)))
    if rms < target_rms:
        audio = audio * target_rms / rms
    if sr != target_sample_rate:
        resampler = torchaudio.transforms.Resample(sr, target_sample_rate)
        audio = resampler(audio)
    audio = audio.to(device)

    # Process each chunk
    results = []
    spectrograms = []

    for i, chunk in enumerate(chunks):
        gr.Info(f"Processing chunk {i+1}/{len(chunks)}: {chunk[:30]}...")

        # Prepare the text
        text_list = [ref_text + chunk]
        final_text_list = convert_char_to_pinyin(text_list)

        # Calculate duration
        ref_audio_len = audio.shape[-1] // hop_length
        zh_pause_punc = r"。，、；：？！"
        ref_text_len = len(ref_text) + len(re.findall(zh_pause_punc, ref_text))
        gen_text_len = len(chunk) + len(re.findall(zh_pause_punc, chunk))
        duration = ref_audio_len + int(ref_audio_len / ref_text_len * gen_text_len / speed)

        # Inference
        gr.Info(f"Generating audio using {exp_name}")
        with torch.inference_mode():
            generated, _ = ema_model.sample(
                cond=audio,
                text=final_text_list,
                duration=duration,
                steps=nfe_step,
                cfg_strength=cfg_strength,
                sway_sampling_coef=sway_sampling_coef,
            )

        generated = generated[:, ref_audio_len:, :]
        generated_mel_spec = rearrange(generated, '1 n d -> 1 d n')

        # Clear unnecessary tensors
        del generated
        devicetorch.empty_cache(torch)

        gr.Info("Running vocoder")
        vocos = Vocos.from_pretrained("charactr/vocos-mel-24khz")
        generated_wave = vocos.decode(generated_mel_spec.cpu())
        if rms < target_rms:
            generated_wave = generated_wave * rms / target_rms

        # Convert to numpy and clear GPU tensors
        generated_wave = generated_wave.squeeze().cpu().numpy()
        del generated_mel_spec
        devicetorch.empty_cache(torch)

        results.append(generated_wave)

        # Generate spectrogram
        with tempfile.NamedTemporaryFile(suffix=".png", delete=False) as tmp_spectrogram:
            spectrogram_path = tmp_spectrogram.name
            save_spectrogram(generated_wave, target_sample_rate, spectrogram_path)
        spectrograms.append(spectrogram_path)

        # Clear cache after processing each chunk
        gc.collect()
        devicetorch.empty_cache(torch)

    # Combine all audio chunks
    combined_audio = np.concatenate(results)

    if remove_silence:
        gr.Info("Removing audio silences... This may take a moment")
        non_silent_intervals = librosa.effects.split(combined_audio, top_db=30)
        non_silent_wave = np.array([])
        for interval in non_silent_intervals:
            start, end = interval
            non_silent_wave = np.concatenate([non_silent_wave, combined_audio[start:end]])
        combined_audio = non_silent_wave

    # Generate final spectrogram
    with tempfile.NamedTemporaryFile(suffix=".png", delete=False) as tmp_spectrogram:
        final_spectrogram_path = tmp_spectrogram.name
        save_spectrogram(combined_audio, target_sample_rate, final_spectrogram_path)

    # Final cleanup
    gc.collect()
    devicetorch.empty_cache(torch)

    # Return combined audio and the final spectrogram
    return (target_sample_rate, combined_audio), final_spectrogram_path

with gr.Blocks() as app:
    ref_audio_input = gr.Audio(label="Reference Audio", type="filepath")
    gen_text_input = gr.Textbox(label="Text to Generate (for longer than 200 chars the app uses chunking)", lines=4)
    model_choice = gr.Radio(choices=["F5-TTS", "E2-TTS"], label="Choose TTS Model", value="F5-TTS")
    generate_btn = gr.Button("Synthesize", variant="primary")
    with gr.Accordion("Advanced Settings", open=False):
        ref_text_input = gr.Textbox(label="Reference Text", info="Leave blank to automatically transcribe the reference audio. If you enter text it will override automatic transcription.", lines=2)
        remove_silence = gr.Checkbox(label="Remove Silences", info="The model tends to produce silences, especially on longer audio. We can manually remove silences if needed. Note that this is an experimental feature and may produce strange results. This will also increase generation time.", value=True)

    audio_output = gr.Audio(label="Synthesized Audio")
    spectrogram_output = gr.Image(label="Spectrogram")

    generate_btn.click(infer, inputs=[ref_audio_input, ref_text_input, gen_text_input, model_choice, remove_silence], outputs=[audio_output, spectrogram_output])
    gr.Markdown("Unofficial demo by [mrfakename](https://x.com/realmrfakename)")


app.queue().launch()