app.py

from huggingface_hub import snapshot_download
from katsu import Katsu
from models import build_model
import gradio as gr
import numpy as np
import os
import phonemizer
import pypdf
import random
import re
import spaces
import torch
import yaml

CUDA_AVAILABLE = torch.cuda.is_available()

snapshot = snapshot_download(repo_id='hexgrad/kokoro', allow_patterns=['*.pt', '*.pth', '*.yml'], use_auth_token=os.environ['TOKEN'])
config = yaml.safe_load(open(os.path.join(snapshot, 'config.yml')))

models = {device: build_model(config['model_params'], device) for device in ['cpu'] + (['cuda'] if CUDA_AVAILABLE else [])}
for key, state_dict in torch.load(os.path.join(snapshot, 'net.pth'), map_location='cpu', weights_only=True)['net'].items():
    for device in models:
        assert key in models[device], key
        try:
            models[device][key].load_state_dict(state_dict)
        except:
            state_dict = {k[7:]: v for k, v in state_dict.items()}
            models[device][key].load_state_dict(state_dict, strict=False)

PARAM_COUNT = sum(p.numel() for value in models['cpu'].values() for p in value.parameters())
assert PARAM_COUNT < 82_000_000, PARAM_COUNT

random_texts = {}
for lang in ['en', 'ja']:
    with open(f'{lang}.txt', 'r') as r:
        random_texts[lang] = [line.strip() for line in r]

def get_random_text(voice):
    if voice[0] == 'j':
        lang = 'ja'
    else:
        lang = 'en'
    return random.choice(random_texts[lang])

def parens_to_angles(s):
    return s.replace('(', '«').replace(')', '»')

def split_num(num):
    num = num.group()
    if '.' in num:
        # Decimal
        a, b = num.split('.')
        return ' point '.join([a, ' '.join(b)])
    elif ':' in num:
        # Time
        h, m = [int(n) for n in num.split(':')]
        if m == 0:
            return f"{h} o'clock"
        elif m < 10:
            return f'{h} oh {m}'
        return f'{h} {m}'
    # Year
    year = int(num[:4])
    if year < 1100 or year % 1000 < 10:
        return num
    left, right = num[:2], int(num[2:4])
    s = 's' if num.endswith('s') else ''
    if 100 <= year % 1000 <= 999:
        if right == 0:
            return f'{left} hundred{s}'
        elif right < 10:
            return f'{left} oh {right}{s}'
    return f'{left} {right}{s}'

def normalize(text):
    # TODO: Custom text normalization rules?
    text = re.sub(r'\bD[Rr]\.(?= [A-Z])', 'Doctor', text)
    text = re.sub(r'\b(?:Mr\.|MR\.(?= [A-Z]))', 'Mister', text)
    text = re.sub(r'\b(?:Ms\.|MS\.(?= [A-Z]))', 'Miss', text)
    text = re.sub(r'\b(?:Mrs\.|MRS\.(?= [A-Z]))', 'Mrs', text)
    text = re.sub(r'\betc\.(?! [A-Z])', 'etc', text)
    text = re.sub(r'\b([Yy])eah\b', r"\1e'a", text)
    text = text.replace(chr(8216), "'").replace(chr(8217), "'")
    text = text.replace(chr(8220), '"').replace(chr(8221), '"')
    text = re.sub(r'[^\S \n]', ' ', text)
    text = re.sub(r'  +', ' ', text)
    text = re.sub(r'(?<=\n) +(?=\n)', '', text)
    text = re.sub(r'\d*\.\d+|\b\d{4}s?\b|(?<!:)\b(?:[1-9]|1[0-2]):[0-5]\d\b(?!:)', split_num, text)
    text = re.sub(r'(?<=\d),(?=\d)', '', text)
    text = re.sub(r'(?<=\d)-(?=\d)', ' to ', text) # TODO: could be minus
    text = re.sub(r'(?<=\d)S', ' S', text)
    text = re.sub(r"(?<=[A-Z])'?s", lambda m: m.group().upper(), text)
    text = re.sub(r'(?:[A-Za-z]\.){2,} [a-z]', lambda m: m.group().replace('.', '-'), text)
    text = re.sub(r'(?i)(?<=[A-Z])\.(?=[A-Z])', '-', text)
    return parens_to_angles(text).strip()

phonemizers = dict(
    a=phonemizer.backend.EspeakBackend(language='en-us', preserve_punctuation=True, with_stress=True),
    b=phonemizer.backend.EspeakBackend(language='en-gb', preserve_punctuation=True, with_stress=True),
    j=Katsu(),
)

def phonemize(text, voice, norm=True):
    lang = voice[0]
    if norm:
        text = normalize(text)
    ps = phonemizers[lang].phonemize([text])
    ps = ps[0] if ps else ''
    # TODO: Custom phonemization rules?
    ps = parens_to_angles(ps)
    # https://en.wiktionary.org/wiki/kokoro#English
    if lang in 'ab':
        ps = ps.replace('kəkˈoːɹoʊ', 'kˈoʊkəɹoʊ').replace('kəkˈɔːɹəʊ', 'kˈəʊkəɹəʊ')
        ps = ps.replace('ʲ', 'j').replace('r', 'ɹ').replace('x', 'k').replace('ɬ', 'l')
        ps = re.sub(r'(?<=[a-zɹː])(?=hˈʌndɹɪd)', ' ', ps)
        ps = re.sub(r' z(?=[;:,.!?¡¿—…"«»“” ]|$)', 'z', ps)
        if lang == 'a':
            ps = re.sub(r'(?<=nˈaɪn)ti(?!ː)', 'di', ps)
    ps = ''.join(filter(lambda p: p in VOCAB, ps))
    if lang == 'j' and any(p in 'ABCDEFGHIJKLMNOPQRSTUVWXYZ' for p in ps):
        gr.Warning('Japanese tokenizer does not handle English letters')
    return ps.strip()

def length_to_mask(lengths):
    mask = torch.arange(lengths.max()).unsqueeze(0).expand(lengths.shape[0], -1).type_as(lengths)
    mask = torch.gt(mask+1, lengths.unsqueeze(1))
    return mask

def get_vocab():
    _pad = "$"
    _punctuation = ';:,.!?¡¿—…"«»“” '
    _letters = 'ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz'
    _letters_ipa = "ɑɐɒæɓʙβɔɕçɗɖðʤəɘɚɛɜɝɞɟʄɡɠɢʛɦɧħɥʜɨɪʝɭɬɫɮʟɱɯɰŋɳɲɴøɵɸθœɶʘɹɺɾɻʀʁɽʂʃʈʧʉʊʋⱱʌɣɤʍχʎʏʑʐʒʔʡʕʢǀǁǂǃˈˌːˑʼʴʰʱʲʷˠˤ˞↓↑→↗↘'̩'ᵻ"
    symbols = [_pad] + list(_punctuation) + list(_letters) + list(_letters_ipa)
    dicts = {}
    for i in range(len((symbols))):
        dicts[symbols[i]] = i
    return dicts

VOCAB = get_vocab()

def tokenize(ps):
    return [i for i in map(VOCAB.get, ps) if i is not None]

# ⭐ voices are stable, 🧪 voices are unstable
CHOICES = {
'🇺🇸 🚺 American Female ⭐': 'af',
'🇺🇸 🚺 Bella': 'af_bella',
'🇺🇸 🚺 Nicole': 'af_nicole',
'🇺🇸 🚺 Sarah': 'af_sarah',
'🇺🇸 🚺 Sky 🧪': 'af_sky',
'🇺🇸 🚹 Adam 🧪': 'am_adam',
'🇺🇸 🚹 Michael': 'am_michael',
'🇬🇧 🚹 Lewis 🧪': 'bm_lewis',
'🇯🇵 🚺 Japanese Female': 'jf_0',
}
VOICES = {device: {k: torch.load(os.path.join(snapshot, 'voicepacks', f'{k}.pt'), weights_only=True).to(device) for k in CHOICES.values()} for device in models}

np_log_99 = np.log(99)
def s_curve(p):
    if p <= 0:
        return 0
    elif p >= 1:
        return 1
    s = 1 / (1 + np.exp((1-p*2)*np_log_99))
    s = (s-0.01) * 50/49
    return s

SAMPLE_RATE = 24000

@torch.no_grad()
def forward(tokens, voice, speed, device='cpu'):
    ref_s = VOICES[device][voice][len(tokens)]
    tokens = torch.LongTensor([[0, *tokens, 0]]).to(device)
    input_lengths = torch.LongTensor([tokens.shape[-1]]).to(device)
    text_mask = length_to_mask(input_lengths).to(device)
    bert_dur = models[device].bert(tokens, attention_mask=(~text_mask).int())
    d_en = models[device].bert_encoder(bert_dur).transpose(-1, -2)
    s = ref_s[:, 128:]
    d = models[device].predictor.text_encoder(d_en, s, input_lengths, text_mask)
    x, _ = models[device].predictor.lstm(d)
    duration = models[device].predictor.duration_proj(x)
    duration = torch.sigmoid(duration).sum(axis=-1) / speed
    pred_dur = torch.round(duration).clamp(min=1).long()
    pred_aln_trg = torch.zeros(input_lengths, pred_dur.sum().item())
    c_frame = 0
    for i in range(pred_aln_trg.size(0)):
        pred_aln_trg[i, c_frame:c_frame + pred_dur[0,i].item()] = 1
        c_frame += pred_dur[0,i].item()
    en = d.transpose(-1, -2) @ pred_aln_trg.unsqueeze(0).to(device)
    F0_pred, N_pred = models[device].predictor.F0Ntrain(en, s)
    t_en = models[device].text_encoder(tokens, input_lengths, text_mask)
    asr = t_en @ pred_aln_trg.unsqueeze(0).to(device)
    return models[device].decoder(asr, F0_pred, N_pred, ref_s[:, :128]).squeeze().cpu().numpy()

@spaces.GPU(duration=10)
def forward_gpu(tokens, voice, speed):
    return forward(tokens, voice, speed, device='cuda')

# Must be backwards compatible with https://huggingface.co./spaces/Pendrokar/TTS-Spaces-Arena
def generate(text, voice, ps, speed, _reduce_noise, opening_cut, closing_cut, ease_in, ease_out, _pad_before, _pad_after, use_gpu):
    return _generate(text, voice, ps, speed, opening_cut, closing_cut, ease_in, ease_out, use_gpu)

def _generate(text, voice, ps, speed, opening_cut, closing_cut, ease_in, ease_out, use_gpu):
    if voice not in VOICES['cpu']:
        voice = 'af'
    ps = ps or phonemize(text, voice)
    tokens = tokenize(ps)
    if not tokens:
        return (None, '')
    elif len(tokens) > 510:
        tokens = tokens[:510]
    ps = ''.join(next(k for k, v in VOCAB.items() if i == v) for i in tokens)
    try:
        if not use_gpu or (use_gpu == 'auto' and len(ps) < 100):
            out = forward(tokens, voice, speed)
        else:
            out = forward_gpu(tokens, voice, speed)
    except gr.exceptions.Error as e:
        raise gr.Error(e)
        return (None, '')
    opening_cut = int(opening_cut / speed)
    if opening_cut > 0:
        out = out[opening_cut:]
    closing_cut = int(closing_cut / speed)
    if closing_cut > 0:
        out = out[:-closing_cut]
    ease_in = min(int(ease_in / speed), len(out)//2)
    for i in range(ease_in):
        out[i] *= s_curve(i / ease_in)
    ease_out = min(int(ease_out / speed), len(out)//2)
    for i in range(ease_out):
        out[-i-1] *= s_curve(i / ease_out)
    return ((SAMPLE_RATE, out), ps)

def toggle_autoplay(autoplay):
    return gr.Audio(interactive=False, label='Output Audio', autoplay=autoplay)

USE_GPU_CHOICES = [('Auto', 'auto'), ('CPU', False), ('ZeroGPU', True)]
USE_GPU_INFOS = {
    'auto': 'Use CPU or GPU, whichever is faster',
    False: 'CPU is ~faster <100 tokens',
    True: 'ZeroGPU is ~faster >100 tokens',
}
def change_use_gpu(value):
    return gr.Dropdown(USE_GPU_CHOICES, value=value, label='Hardware', info=USE_GPU_INFOS[value], interactive=CUDA_AVAILABLE)

with gr.Blocks() as basic_tts:
    with gr.Row():
        with gr.Column():
            text = gr.Textbox(label='Input Text', info='Generate speech for one segment of text using Kokoro, a TTS model with 80 million parameters')
            with gr.Row():
                voice = gr.Dropdown(list(CHOICES.items()), value='af', label='Voice', info='⭐ voices are stable, 🧪 voices are unstable')
                use_gpu = gr.Dropdown(
                    USE_GPU_CHOICES,
                    value='auto' if CUDA_AVAILABLE else False,
                    label='Hardware',
                    info=USE_GPU_INFOS['auto' if CUDA_AVAILABLE else False],
                    interactive=CUDA_AVAILABLE
                )
                use_gpu.change(fn=change_use_gpu, inputs=[use_gpu], outputs=[use_gpu])
            with gr.Row():
                random_btn = gr.Button('Random Text', variant='secondary')
                generate_btn = gr.Button('Generate', variant='primary')
            random_btn.click(get_random_text, inputs=[voice], outputs=[text])
            with gr.Accordion('Input Tokens', open=False):
                in_ps = gr.Textbox(show_label=False, info='Override the input text with custom phonemes. Leave this blank to automatically tokenize the input text instead.')
                with gr.Row():
                    clear_btn = gr.ClearButton(in_ps)
                    phonemize_btn = gr.Button('Tokenize Input Text', variant='primary')
            phonemize_btn.click(phonemize, inputs=[text, voice], outputs=[in_ps])
        with gr.Column():
            audio = gr.Audio(interactive=False, label='Output Audio', autoplay=True)
            autoplay = gr.Checkbox(value=True, label='Autoplay')
            autoplay.change(toggle_autoplay, inputs=[autoplay], outputs=[audio])
            with gr.Accordion('Output Tokens', open=True):
                out_ps = gr.Textbox(interactive=False, show_label=False, info='Tokens used to generate the audio, up to 510 allowed. Same as input tokens if supplied, excluding unknowns.')
    with gr.Accordion('Audio Settings', open=False):
        with gr.Row():
            speed = gr.Slider(minimum=0.5, maximum=2, value=1, step=0.1, label='⚡️ Speed', info='Adjust the speed of the audio; the settings below are auto-scaled by speed')
        with gr.Row():
            with gr.Column():
                opening_cut = gr.Slider(minimum=0, maximum=24000, value=4000, step=1000, label='✂️ Opening Cut', info='Cut samples from the start')
            with gr.Column():
                closing_cut = gr.Slider(minimum=0, maximum=24000, value=2000, step=1000, label='🎬 Closing Cut', info='Cut samples from the end')
        with gr.Row():
            with gr.Column():
                ease_in = gr.Slider(minimum=0, maximum=24000, value=3000, step=1000, label='🎢 Ease In', info='Ease in samples, after opening cut')
            with gr.Column():
                ease_out = gr.Slider(minimum=0, maximum=24000, value=1000, step=1000, label='🛝 Ease Out', info='Ease out samples, before closing cut')
    text.submit(_generate, inputs=[text, voice, in_ps, speed, opening_cut, closing_cut, ease_in, ease_out, use_gpu], outputs=[audio, out_ps])
    generate_btn.click(_generate, inputs=[text, voice, in_ps, speed, opening_cut, closing_cut, ease_in, ease_out, use_gpu], outputs=[audio, out_ps])

@torch.no_grad()
def lf_forward(token_lists, voice, speed, device='cpu'):
    voicepack = VOICES[device][voice]
    outs = []
    for tokens in token_lists:
        ref_s = voicepack[len(tokens)]
        s = ref_s[:, 128:]
        tokens = torch.LongTensor([[0, *tokens, 0]]).to(device)
        input_lengths = torch.LongTensor([tokens.shape[-1]]).to(device)
        text_mask = length_to_mask(input_lengths).to(device)
        bert_dur = models[device].bert(tokens, attention_mask=(~text_mask).int())
        d_en = models[device].bert_encoder(bert_dur).transpose(-1, -2)
        d = models[device].predictor.text_encoder(d_en, s, input_lengths, text_mask)
        x, _ = models[device].predictor.lstm(d)
        duration = models[device].predictor.duration_proj(x)
        duration = torch.sigmoid(duration).sum(axis=-1) / speed
        pred_dur = torch.round(duration).clamp(min=1).long()
        pred_aln_trg = torch.zeros(input_lengths, pred_dur.sum().item())
        c_frame = 0
        for i in range(pred_aln_trg.size(0)):
            pred_aln_trg[i, c_frame:c_frame + pred_dur[0,i].item()] = 1
            c_frame += pred_dur[0,i].item()
        en = d.transpose(-1, -2) @ pred_aln_trg.unsqueeze(0).to(device)
        F0_pred, N_pred = models[device].predictor.F0Ntrain(en, s)
        t_en = models[device].text_encoder(tokens, input_lengths, text_mask)
        asr = t_en @ pred_aln_trg.unsqueeze(0).to(device)
        outs.append(models[device].decoder(asr, F0_pred, N_pred, ref_s[:, :128]).squeeze().cpu().numpy())
    return outs

@spaces.GPU
def lf_forward_gpu(token_lists, voice, speed):
    return lf_forward(token_lists, voice, speed, device='cuda')

def resplit_strings(arr):
    # Handle edge cases
    if not arr:
        return '', ''
    if len(arr) == 1:
        return arr[0], ''
    # Try each possible split point
    min_diff = float('inf')
    best_split = 0
    # Calculate lengths when joined with spaces
    lengths = [len(s) for s in arr]
    spaces = len(arr) - 1  # Total spaces needed
    # Try each split point
    left_len = 0
    right_len = sum(lengths) + spaces
    for i in range(1, len(arr)):
        # Add current word and space to left side
        left_len += lengths[i-1] + (1 if i > 1 else 0)
        # Remove current word and space from right side
        right_len -= lengths[i-1] + 1
        diff = abs(left_len - right_len)
        if diff < min_diff:
            min_diff = diff
            best_split = i
    # Join the strings with the best split point
    return ' '.join(arr[:best_split]), ' '.join(arr[best_split:])

def recursive_split(text, voice):
    if not text:
        return []
    tokens = phonemize(text, voice, norm=False)
    if len(tokens) < 511:
        return [(text, tokens, len(tokens))] if tokens else []
    if ' ' not in text:
        return []
    for punctuation in ['!.?…', ':;', ',—']:
        splits = re.split(f'(?:(?<=[{punctuation}])|(?<=[{punctuation}]["\'»])|(?<=[{punctuation}]["\'»]["\'»])) ', text)
        if len(splits) > 1:
            break
        else:
            splits = None
    splits = splits or text.split(' ')
    a, b = resplit_strings(splits)
    return recursive_split(a, voice) + recursive_split(b, voice)

def segment_and_tokenize(text, voice, skip_square_brackets=True, newline_split=2):
    if skip_square_brackets:
        text = re.sub(r'\[.*?\]', '', text)
    texts = [t.strip() for t in re.split('\n{'+str(newline_split)+',}', normalize(text))] if newline_split > 0 else [normalize(text)]
    segments = [row for t in texts for row in recursive_split(t, voice)]
    return [(i, *row) for i, row in enumerate(segments)]

def lf_generate(segments, voice, speed, opening_cut, closing_cut, ease_in, ease_out, pad_between, use_gpu):
    token_lists = list(map(tokenize, segments['Tokens']))
    wavs = []
    opening_cut = int(opening_cut / speed)
    closing_cut = int(closing_cut / speed)
    pad_between = int(pad_between / speed)
    batch_size = 100
    for i in range(0, len(token_lists), batch_size):
        try:
            if use_gpu:
                outs = lf_forward_gpu(token_lists[i:i+batch_size], voice, speed)
            else:
                outs = lf_forward(token_lists[i:i+batch_size], voice, speed)
        except gr.exceptions.Error as e:
            if wavs:
                gr.Warning(str(e))
            else:
                raise gr.Error(e)
            break
        for out in outs:
            if opening_cut > 0:
                out = out[opening_cut:]
            if closing_cut > 0:
                out = out[:-closing_cut]
            ease_in = min(int(ease_in / speed), len(out)//2)
            for i in range(ease_in):
                out[i] *= s_curve(i / ease_in)
            ease_out = min(int(ease_out / speed), len(out)//2)
            for i in range(ease_out):
                out[-i-1] *= s_curve(i / ease_out)
            if wavs and pad_between > 0:
                wavs.append(np.zeros(pad_between))
            wavs.append(out)
    return (SAMPLE_RATE, np.concatenate(wavs)) if wavs else None

def did_change_segments(segments):
    x = len(segments) if segments['Length'].any() else 0
    return [
        gr.Button('Tokenize', variant='secondary' if x else 'primary'),
        gr.Button(f'Generate x{x}', variant='primary' if x else 'secondary', interactive=x > 0),
    ]

def extract_text(file):
    if file.endswith('.pdf'):
        with open(file, 'rb') as rb:
            pdf_reader = pypdf.PdfReader(rb)
            return '\n'.join([page.extract_text() for page in pdf_reader.pages])
    elif file.endswith('.txt'):
        with open(file, 'r') as r:
            return '\n'.join([line for line in r])
    return None

with gr.Blocks() as lf_tts:
    with gr.Row():
        with gr.Column():
            file_input = gr.File(file_types=['.pdf', '.txt'], label='Input File: pdf or txt')
            text = gr.Textbox(label='Input Text', info='Generate speech in batches of 100 text segments and automatically join them together')
            file_input.upload(fn=extract_text, inputs=[file_input], outputs=[text])
            with gr.Row():
                voice = gr.Dropdown(list(CHOICES.items()), value='af', label='Voice', info='⭐ voices are stable, 🧪 voices are unstable')
                use_gpu = gr.Dropdown(
                    [('ZeroGPU', True), ('CPU', False)],
                    value=CUDA_AVAILABLE,
                    label='Hardware',
                    info='GPU is >10x faster but has a usage quota',
                    interactive=CUDA_AVAILABLE
                )
            with gr.Accordion('Text Settings', open=False):
                skip_square_brackets = gr.Checkbox(True, label='Skip [Square Brackets]', info='Recommended for academic papers, Wikipedia articles, or texts with citations')
                newline_split = gr.Number(2, label='Newline Split', info='Split the input text on this many newlines. Affects how the text is segmented.', precision=0, minimum=0)
            with gr.Row():
                segment_btn = gr.Button('Tokenize', variant='primary')
                generate_btn = gr.Button('Generate x0', variant='secondary', interactive=False)
        with gr.Column():
            audio = gr.Audio(interactive=False, label='Output Audio')
            with gr.Accordion('Audio Settings', open=False):
                with gr.Row():
                    speed = gr.Slider(minimum=0.5, maximum=2, value=1, step=0.1, label='⚡️ Speed', info='Adjust the speed of the audio; the settings below are auto-scaled by speed')
                with gr.Row():
                    with gr.Column():
                        opening_cut = gr.Slider(minimum=0, maximum=24000, value=4000, step=1000, label='✂️ Opening Cut', info='Cut samples from the start')
                    with gr.Column():
                        closing_cut = gr.Slider(minimum=0, maximum=24000, value=2000, step=1000, label='🎬 Closing Cut', info='Cut samples from the end')
                with gr.Row():
                    with gr.Column():
                        ease_in = gr.Slider(minimum=0, maximum=24000, value=3000, step=1000, label='🎢 Ease In', info='Ease in samples, after opening cut')
                    with gr.Column():
                        ease_out = gr.Slider(minimum=0, maximum=24000, value=1000, step=1000, label='🛝 Ease Out', info='Ease out samples, before closing cut')
                with gr.Row():
                    pad_between = gr.Slider(minimum=0, maximum=24000, value=10000, step=1000, label='🔇 Pad Between', info='How many samples of silence to insert between segments')
    with gr.Row():
        segments = gr.Dataframe(headers=['#', 'Text', 'Tokens', 'Length'], row_count=(1, 'dynamic'), col_count=(4, 'fixed'), label='Segments', interactive=False, wrap=True)
        segments.change(fn=did_change_segments, inputs=[segments], outputs=[segment_btn, generate_btn])
    segment_btn.click(segment_and_tokenize, inputs=[text, voice, skip_square_brackets, newline_split], outputs=[segments])
    generate_btn.click(lf_generate, inputs=[segments, voice, speed, opening_cut, closing_cut, ease_in, ease_out, pad_between, use_gpu], outputs=[audio])

with gr.Blocks() as about:
    gr.Markdown("""
Kokoro is a frontier TTS model for its size. It has 80 million parameters,<sup>[1]</sup> uses a lean StyleTTS 2 architecture,<sup>[2]</sup> and was trained on high-quality data. The weights are currently private, but a free public demo is hosted here, at `https://hf.co/spaces/hexgrad/Kokoro-TTS`. The Community tab is open for feature requests, bug reports, etc. For other inquiries, contact @rzvzn on Discord.

### FAQ
#### Will this be open sourced?
There currently isn't a release date scheduled for the weights. The inference code in this space is MIT licensed. The architecture was already published by Li et al, with MIT licensed code and pretrained weights.<sup>[2]</sup>

#### What is an unstable voice?
An unstable voice is more likely to stumble or produce unnatural artifacts, especially on short or strange texts.

#### How can CPU be faster than ZeroGPU?
The CPU is a dedicated resource for this Space, while the ZeroGPU pool is shared and dynamically allocated across all of HF. The ZeroGPU queue/allocator system inevitably adds latency to each request.<br/>
For Basic TTS under ~100 tokens or characters, only a few seconds of audio need to be generated, so the actual compute is not that heavy. In these short bursts, the dedicated CPU can often compute the result faster than the total time it takes to: enter the ZeroGPU queue, wait to get allocated, and have a GPU compute and deliver the result.<br/>
ZeroGPU catches up beyond 100 tokens and especially closer to the ~500 token context window. Long-Form mode processes batches of 100 segments at a time, so the GPU should outspeed the CPU by 1-2 orders of magnitude.

### Compute
The model was trained on 1x A100-class 80GB instances rented from [Vast.ai](https://cloud.vast.ai/?ref_id=79907).<sup>[3]</sup><br/>
Vast was chosen over other compute providers due to its competitive on-demand hourly rates.<br/>
The average hourly cost for the 1x A100-class 80GB VRAM instances used for training was below $1/hr — around half the quoted rates from other providers.

### Gradio API
This Space can be used via API. The following code block can be copied and run in one Google Colab cell.
```
# 1. Install the Gradio Python client
!pip install -q gradio_client

# 2. Initialize the client
from gradio_client import Client
client = Client('hexgrad/Kokoro-TTS')

# 3. Call the generate endpoint, which returns a pair: an audio path and a string of output phonemes
audio_path, out_ps = client.predict(
    text="How could I know? It's an unanswerable question. Like asking an unborn child if they'll lead a good life. They haven't even been born.",
    voice='af',
    api_name='/generate'
)

# 4. Display the audio and print the output phonemes
from IPython.display import display, Audio
display(Audio(audio_path, autoplay=True))
print(out_ps)
```
Note that this Space and the underlying Kokoro model are both under development and subject to change. Reliability is not guaranteed. Hugging Face and/or Gradio might enforce their own rate limits.

### Model Version History
| Version | Date | Val mel / dur / f0 Losses |
| ------- | ---- | ------------------------- |
| v0.19 | 2024 Nov 22 | 0.261 / 0.627 / 1.897 |
| v0.16 | 2024 Nov 15 | 0.263 / 0.646 / 1.934 |
| v0.14 | 2024 Nov 12 | 0.262 / 0.642 / 1.889 |

### Licenses
Inference code: MIT<br/>
espeak-ng dependency: GPL-3.0<sup>[4]</sup><br/>
Random English texts: Unknown<sup>[5]</sup><br/>
Random Japanese texts: CC0 public domain<sup>[6]</sup>

### References
1. Kokoro parameter count | https://hf.co/spaces/hexgrad/Kokoro-TTS/blob/main/app.py#L31
2. StyleTTS 2 | https://github.com/yl4579/StyleTTS2
3. Vast.ai referral link | https://cloud.vast.ai/?ref_id=79907
4. eSpeak NG | https://github.com/espeak-ng/espeak-ng
5. Quotable Data | https://github.com/quotable-io/data/blob/master/data/quotes.json
6. Common Voice Japanese sentences | https://github.com/common-voice/common-voice/tree/main/server/data/ja
""")

with gr.Blocks() as app:
    gr.TabbedInterface(
        [basic_tts, lf_tts, about],
        ['🔥 Basic TTS', '📖 Long-Form', 'ℹ️ About'],
    )

if __name__ == '__main__':
    app.queue(api_open=True).launch()