Hugging Face's logo

Spaces:
Dubverse
/
MahaTTSv2
Runtime error

App Files Community
MahaTTSv2
File size: 3,153 Bytes
2223778
 
 
 
 
 
 
 
 
 
 
 
 
 
184cb2e
2223778
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
bd7d901
 
 
2223778
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
184cb2e
2223778
 
 
 
 
 
 
 
 
 
 
 
184cb2e
2223778
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
 
#setting up
import spaces
import sys, os, torch
import subprocess

token = os.environ["TOKEN"]
repo = "Dubverse/MahaTTSv2"
clone_dir = "MahaTTSv2"
url = f"https://__token__:{token}:@huggingface.co/{repo}"  # NOTE the ':' after token
subprocess.run("git lfs install",shell=True)
subprocess.run(f"git clone --recurse-submodules {url}",shell=True)


sys.path.append("MahaTTSv2/")
import gradio as gr
from inference import infer, prepare_inputs, load_t2s_model, load_cfm, create_wav_header
from tqdm import tqdm

# Setup
os.makedirs("generated_samples/", exist_ok=True)
device = "cuda"# if torch.cuda.is_available() else "cpu"
print("Using device", device)

# Model checkpoints
m1_checkpoint = "MahaTTSv2/pretrained_checkpoint/m1_gemma_benchmark_1_latest_weights.pt"
m2_checkpoint = "MahaTTSv2/pretrained_checkpoint/m2.pt"
vocoder_checkpoint = 'MahaTTSv2/pretrained_checkpoint/700_580k_multilingual_infer_ready/'

global FM, vocoder, m2, mu, std, m1

# Load models
FM, vocoder, m2, mu, std = load_cfm(m2_checkpoint, vocoder_checkpoint, device)
m1 = load_t2s_model(m1_checkpoint, device)


# Speaker reference clips
speaker_refs = {
    "Female_Speaker1": [
            "MahaTTSv2/speakers/female1/train_hindifemale_02794.wav",
            "MahaTTSv2/speakers/female1/train_hindifemale_04167.wav",
            "MahaTTSv2/speakers/female1/train_hindifemale_02795.wav"
        ],
    "Male_Speaker1": [
            "MahaTTSv2/speakers/male1/train_hindimale_00016.wav",
            "MahaTTSv2/speakers/male1/train_hindimale_00017.wav",
            "MahaTTSv2/speakers/male1/train_hindimale_00018.wav"
        ]
}

# Available languages (can be extended)
available_languages = [
        "assamese",
        "bengali",
        "bhojpuri",
        "bodo",
        "dogri",
        "odia",
        "english",
        "french",
        "gujarati",
        "german",
        "hindi",
        "italian",
        "kannada",
        "malayalam",
        "marathi",
        "punjabi",
        "rajasthani",
        "sanskrit",
        "spanish",
        "tamil",
        "telugu",
    ]

# Inference function
@spaces.GPU(duration=60)
def generate_audio(text, speaker_name, language):
    if speaker_name not in speaker_refs:
        return f"Reference clips not available for {speaker_name}", None

    ref_clips = speaker_refs[speaker_name]   

    text_ids, code_ids, language_code, ref_mels_m1, ref_mels_m2 = prepare_inputs(
        text.lower(),
        ref_clips_m1=ref_clips,
        ref_clips_m2=ref_clips,
        language=language,
        device=device
    )

    audio_wav = infer(m1, m2, vocoder, FM, mu, std, text_ids, code_ids, language_code, ref_mels_m1, ref_mels_m2, device)
    return 24000,audio_wav

# Gradio UI
interface = gr.Interface(
    fn=generate_audio,
    inputs=[
        gr.Textbox(label="Enter Text"),
        gr.Dropdown(choices=list(speaker_refs.keys()), label="Select Speaker"),
        gr.Dropdown(choices=available_languages, label="Select Language")
    ],
    outputs=gr.Audio(label="Generated Speech"),
    title="MahaTTSv2 Demo",
    description="Enter text, choose a speaker and language to generate speech."
)

interface.launch()