Merge branch 'aya' of https://huggingface.co/spaces/AiDi-UIR/TTS-Conv-Darija

app.py

@@ -0,0 +1,80 @@
+import gradio as gr
+from huggingface_hub import InferenceClient
+from gradio_client import Client, handle_file
+import tempfile
+import os
+
+# Load TTS client - using the correct Space URL
+tts_client = Client("https://medmac01-darija-arabic-tts.hf.space/")
+
+# Load text generation client
+text_client = InferenceClient("deepseek-ai/DeepSeek-R1-Distill-Qwen-32B")
+
+def generate_conversation(subject, speaker1_audio, speaker2_audio):
+    prompt = f"""
+    Generate a natural Moroccan Darija conversation between two people about: "{subject}".
+    Format:
+    Speaker 1: ...
+    Speaker 2: ...
+    Speaker 1: ...
+    Speaker 2: ...
+    Keep it short and casual (4 lines).
+    """
+
+    result = text_client.text_generation(prompt, max_new_tokens=300, temperature=0.7)
+    lines = [line.strip() for line in result.split('\n') if line.strip().startswith("Speaker")]
+
+    # Generate audio files using TTS
+    audio_paths = []
+    idx = 0
+    for line in lines:
+        speaker_audio = speaker1_audio if line.startswith("Speaker 1") else speaker2_audio
+        text = line.split(":", 1)[1].strip()
+
+        # Create TTS audio using the correct API call
+        result = tts_client.predict(
+            text=text,
+            speaker_audio_path=handle_file(speaker_audio),
+            temperature=0.75,
+            api_name="/infer_EGTTS"
+        )
+        
+        # Save the result to a temporary file
+        with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp:
+            with open(result, "rb") as f:
+                tmp.write(f.read())
+            tmp.flush()
+            audio_paths.append(tmp.name)
+            idx += 1
+
+    # Format the conversation text
+    conversation_text = "\n".join(lines)
+    
+    # Return all outputs in the correct order
+    return [conversation_text] + audio_paths
+
+with gr.Blocks() as demo:
+    gr.Markdown("# 🗣️ Moroccan Darija Conversation Generator")
+    gr.Markdown("Enter a discussion topic and upload 2 speaker voices. We'll generate a Darija conversation!")
+
+    with gr.Row():
+        subject = gr.Textbox(label="Subject of the discussion", placeholder="e.g. Going to the souk")
+    with gr.Row():
+        speaker1 = gr.Audio(label="Speaker 1 Reference (4-5 sec)", type="filepath")
+        speaker2 = gr.Audio(label="Speaker 2 Reference (4-5 sec)", type="filepath")
+
+    btn = gr.Button("🎤 Generate Conversation")
+    
+    # Add text output for the conversation
+    conversation_output = gr.Textbox(label="Generated Conversation", lines=6)
+    
+    # Audio outputs
+    audio_outputs = [gr.Audio(label=f"Line {i+1}") for i in range(4)]
+
+    btn.click(
+        generate_conversation, 
+        inputs=[subject, speaker1, speaker2], 
+        outputs=[conversation_output] + audio_outputs
+    )
+
+demo.launch()