Initial: initial commit

app.py

@@ -0,0 +1,24 @@
+import gradio as gr
+from tts import synthesize, TTS_EXAMPLES
+
+mms_synthesize = gr.Interface(
+    fn=synthesize,
+    inputs=[
+        gr.Dropdown(["original", "nova", "homhom"], label="Model", value="nova"),
+        gr.Textbox(label="Input text"),
+        gr.Slider(minimum=0.1, maximum=2.0, value=1.0, step=0.1, label="Speed"),
+    ],
+    outputs=[
+        gr.Audio(label="Generated Audio", type="numpy"),
+        gr.Textbox(label="Filtered text after processing"),
+    ],
+    examples=TTS_EXAMPLES,
+    title="Text-to-Speech Demo",
+    description="Generate audio in your desired language from input text.",
+    allow_flagging="never",
+)
+
+with gr.Blocks() as demo:
+    mms_synthesize.render()
+
+demo.launch()

requirements.txt

@@ -0,0 +1,5 @@
+gradio
+transformers
+torch
+torchaudio
+git+https://github.com/NoerNova/ShanNLP

tts.py

@@ -0,0 +1,76 @@
+import os
+from transformers import VitsModel, VitsTokenizer
+import torch
+from shannlp import util, word_tokenize
+
+
+def preprocess_string(input_string: str):
+    input_string = input_string.replace("(", "").replace(")", "")
+    string_token = word_tokenize(input_string)
+    num_to_shanword = util.num_to_shanword
+
+    result = []
+    for token in string_token:
+        if token.strip().isdigit():
+            result.append(num_to_shanword(int(token)))
+        else:
+            result.append(token)
+
+    full_token = "".join(result)
+    return full_token
+
+
+def synthesize(model: str, input_string: str, speed: float = 1.0):
+    auth_token = os.environ.get("TOKEN_READ_SECRET") or True
+
+    model_id = {
+        "original": "facebook/mms-tts-shn",
+        "nova": "NorHsangPha/mms-tts-nova-train",
+        "homhom": "NorHsangPha/mms-tts-shn-train",
+    }[model]
+
+    model = VitsModel.from_pretrained(model_id, token=auth_token)
+    tokenizer = VitsTokenizer.from_pretrained(model_id, token=auth_token)
+
+    device = ""
+    if torch.cuda.is_available():
+        device = torch.device("cuda")
+    elif torch.backends.mps.is_available():
+        device = torch.device("mps")
+    else:
+        device = torch.device("cpu")
+
+    model.to(device)
+
+    processed_string = preprocess_string(input_string)
+    inputs = tokenizer(processed_string, return_tensors="pt").to(device)
+
+    torch.manual_seed(42)
+
+    model.speaking_rate = speed
+    model.noise_scale = 0.2
+    model.noise_scale_w = 0.2
+    model.length_scale = 1.0 / speed
+
+    with torch.no_grad():
+        output = model(**inputs).waveform
+
+    output = output.squeeze().cpu().numpy()
+
+    return ((16_000, output), processed_string)
+
+
+TTS_EXAMPLES = [
+    ["nova", "မႂ်ႇသုင်ၶႃႈ ယူႇလီၵိၼ်ဝၢၼ်ၵတ်းယဵၼ် လီယူႇၶႃႈၼေႃႈ။", 1.0],
+    ["original", "ပဵၼ်ယၢမ်းဢၼ် ၸႂ်တိုၼ်ႇတဵၼ်ႈ ၽူင်ႉပိဝ် တႃႇၼုမ်ႇယိင်းၼုမ်ႇၸၢႆးၶဝ် ၸိူဝ်းဢၼ် တေလႆႈၶိုၼ်ႈႁဵၼ်းၼၼ်ႉယူႇ", 1.0],
+    [
+        "homhom",
+        "မိူဝ်ႈပီ 1958 လိူၼ်မေႊ 21 ဝၼ်းၼၼ်ႉ ၸဝ်ႈၼွႆႉသေႃးယၼ်ႇတ ဢမ်ႇၼၼ် ၸဝ်ႈၼွႆႉ ဢွၼ်ႁူဝ် ၽူႈႁၵ်ႉၸိူဝ်ႉၸၢတ်ႈ 31 ၵေႃႉသေ တိူင်ႇၵၢဝ်ႇယၼ်ႇၸႂ် ၵိၼ်ၼမ်ႉသတ်ႉၸႃႇ တႃႇၵေႃႇတင်ႈပူၵ်းပွင် ၵၢၼ်လုၵ်ႉၽိုၼ်ႉ တီႈႁူၺ်ႈပူႉ ႁိမ်းသူပ်းၼမ်ႉၵျွတ်ႈ ၼႂ်းဢိူင်ႇမိူင်းႁၢင် ၸႄႈဝဵင်းမိူင်းတူၼ် ၸိုင်ႈတႆးပွတ်းဢွၵ်ႇၶူင်း လႅၼ်လိၼ်ၸိုင်ႈထႆး။",
+        1.0,
+    ],
+    [
+        "nova",
+        "ပဵၼ်ၵၢၼ်ၾုၵ်ႇၾင်ၸႂ်ၵၼ်ၼႅၼ်ႈ ၼၵ်းပၵ်းၸႂ် ယွင်ႈၵုၼ်းယွင်ႈမုၼ်ဢူငဝ်း ၸိူဝ်းၽူႈလဵပ်ႈႁဵၼ်းႁူႉပိုၼ်း ၸဵမ်လဵၵ်ႉယႂ်ႇၼုမ်ႇထဝ်ႈ ၼႂ်းၸိူဝ်း ၽူႈႁၵ်ႉ ၸိူဝ်ႉၸၢတ်ႈလၢႆပၢၼ်လၢႆသႅၼ်းမႃး 66 ပီ ၼပ်ႉတင်ႈတႄႇ 1958 ဝၼ်းတီႈ 21 လိူၼ်မေႊ။",
+        1.0,
+    ],
+]