Create app.py

app.py

@@ -0,0 +1,113 @@
+import gradio as gr
+import torch
+import torchaudio
+import re
+import os
+from transformers import SpeechT5Processor, SpeechT5ForTextToSpeech, SpeechT5HifiGan
+from speechbrain.pretrained import EncoderClassifier
+
+device = "cuda" if torch.cuda.is_available() else "cpu"
+
+# Load processor & vocoder
+processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
+vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
+
+# Load both TTS models
+model_male = SpeechT5ForTextToSpeech.from_pretrained("Somalitts/5aad").to(device)
+model_female = SpeechT5ForTextToSpeech.from_pretrained("Somalitts/8aad").to(device)
+
+# Load speaker encoder model
+speaker_model = EncoderClassifier.from_hparams(
+    source="speechbrain/spkrec-xvect-voxceleb",
+    run_opts={"device": device},
+    savedir="./spk_model"
+)
+
+# Auto-generate embedding
+def get_embedding(wav_path, pt_path):
+    if os.path.exists(pt_path):
+        return torch.load(pt_path).to(device)
+    else:
+        audio, sr = torchaudio.load(wav_path)
+        audio = torchaudio.functional.resample(audio, sr, 16000).mean(dim=0).unsqueeze(0).to(device)
+        with torch.no_grad():
+            emb = speaker_model.encode_batch(audio)
+            emb = torch.nn.functional.normalize(emb, dim=2).squeeze()
+        torch.save(emb.cpu(), pt_path)
+        return emb
+
+# Ensure embeddings are created or loaded
+embedding_male = get_embedding("Hussein.wav", "male_embedding.pt")
+embedding_female = get_embedding("caasho.wav", "female_embedding.pt")
+
+# Somali numbers to words
+number_words = {
+    0: "eber", 1: "koow", 2: "labo", 3: "seddex", 4: "afar", 5: "shan",
+    6: "lix", 7: "todobo", 8: "sideed", 9: "sagaal", 10: "toban",
+    11: "toban iyo koow", 12: "toban iyo labo", 13: "toban iyo seddex",
+    14: "toban iyo afar", 15: "toban iyo shan", 16: "toban iyo lix",
+    17: "toban iyo todobo", 18: "toban iyo sideed", 19: "toban iyo sagaal",
+    20: "labaatan", 30: "sodon", 40: "afartan", 50: "konton",
+    60: "lixdan", 70: "todobaatan", 80: "sideetan", 90: "sagaashan",
+    100: "boqol", 1000: "kun",
+}
+
+def number_to_words(number):
+    if number < 20:
+        return number_words[number]
+    elif number < 100:
+        tens, unit = divmod(number, 10)
+        return number_words[tens * 10] + (" " + number_words[unit] if unit else "")
+    elif number < 1000:
+        hundreds, remainder = divmod(number, 100)
+        return (number_words[hundreds] + " boqol" if hundreds > 1 else "BOQOL") + (" " + number_to_words(remainder) if remainder else "")
+    elif number < 1000000:
+        thousands, remainder = divmod(number, 1000)
+        return (number_to_words(thousands) + " kun" if thousands > 1 else "KUN") + (" " + number_to_words(remainder) if remainder else "")
+    elif number < 1000000000:
+        millions, remainder = divmod(number, 1000000)
+        return number_to_words(millions) + " malyan" + (" " + number_to_words(remainder) if remainder else "")
+    elif number < 1000000000000:
+        billions, remainder = divmod(number, 1000000000)
+        return number_to_words(billions) + " milyaar" + (" " + number_to_words(remainder) if remainder else "")
+    else:
+        return str(number)
+
+def replace_numbers_with_words(text):
+    return re.sub(r'\b\d+\b', lambda match: number_to_words(int(match.group())), text)
+
+def normalize_text(text):
+    text = text.lower()
+    text = replace_numbers_with_words(text)
+    text = re.sub(r'[^\w\s]', '', text)
+    return text
+
+# Main TTS function
+def text_to_speech(text, voice):
+    text = normalize_text(text)
+    inputs = processor(text=text, return_tensors="pt").to(device)
+
+    if voice == "Male":
+        model = model_male
+        embedding = embedding_male
+    else:
+        model = model_female
+        embedding = embedding_female
+
+    with torch.no_grad():
+        speech = model.generate_speech(inputs["input_ids"], embedding.unsqueeze(0), vocoder=vocoder)
+    return (16000, speech.cpu().numpy())
+
+# Gradio Interface
+iface = gr.Interface(
+    fn=text_to_speech,
+    inputs=[
+        gr.Textbox(label="Geli qoraalka Af-Soomaaliga", placeholder="Tusaale: Baro aqoonta casriga ah..."),
+        gr.Radio(["Male", "Female"], label="Dooro Codka", value="Female")
+    ],
+    outputs=gr.Audio(label="Codka la abuuray", type="numpy"),
+    title="Somali TTS (Lab & Dhedig)",
+    description="Dooro codka aad rabto, geli qoraal af-soomaali ah, codka ayaa la abuuri doonaa adigoo isticmaalaya Somali TTS (SpeechT5)."
+)
+
+iface.launch()