Create app.py

app.py

@@ -0,0 +1,83 @@
+import gradio as gr
+import torch
+import torchaudio
+from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
+# from huggingface_hub import InferenceClient # Removed
+from ttsmms import download, TTS
+from langdetect import detect
+from gradio_client import Client # Added
+
+# Load ASR Model
+asr_model_name = "Futuresony/Future-sw_ASR-24-02-2025"
+processor = Wav2Vec2Processor.from_pretrained(asr_model_name)
+asr_model = Wav2Vec2ForCTC.from_pretrained(asr_model_name)
+
+# Load Text Generation Model - Using Gradio Client
+# client = InferenceClient("unsloth/gemma-3-1b-it") # Removed
+llm_client = Client("Futuresony/Mr.Events") # Added
+
+# def format_prompt(user_input): # Removed
+#     return f"{user_input}" # Removed
+
+# Load TTS Models
+swahili_dir = download("swh", "./data/swahili")
+english_dir = download("eng", "./data/english")
+
+swahili_tts = TTS(swahili_dir)
+english_tts = TTS(english_dir)
+
+# ASR Function
+def transcribe(audio_file):
+    speech_array, sample_rate = torchaudio.load(audio_file)
+    resampler = torchaudio.transforms.Resample(orig_freq=sample_rate, new_freq=16000)
+    speech_array = resampler(speech_array).squeeze().numpy()
+    input_values = processor(speech_array, sampling_rate=16000, return_tensors="pt").input_values
+    with torch.no_grad():
+        logits = asr_model(input_values).logits
+    predicted_ids = torch.argmax(logits, dim=-1)
+    transcription = processor.batch_decode(predicted_ids)[0]
+    return transcription
+
+# Text Generation Function - Using Gradio Client
+def generate_text(prompt):
+    # formatted_prompt = format_prompt(prompt) # Removed
+    # response = client.text_generation(formatted_prompt, max_new_tokens=250, temperature=0.7, top_p=0.95) # Removed
+    result = llm_client.predict(query=prompt, api_name="/chat") # Added
+    return result.strip() # Modified to return the result from the Gradio Client
+
+# TTS Function
+def text_to_speech(text):
+    lang = detect(text)
+    wav_path = "./output.wav"
+    if lang == "sw":
+        swahili_tts.synthesis(text, wav_path=wav_path)
+    else:
+        english_tts.synthesis(text, wav_path=wav_path)
+    return wav_path
+
+# Combined Processing Function
+def process_audio(audio):
+    transcription = transcribe(audio)
+    generated_text = generate_text(transcription)
+    speech = text_to_speech(generated_text)
+    return transcription, generated_text, speech
+
+# Gradio Interface
+with gr.Blocks() as demo:
+    gr.Markdown("<p align='center' style='font-size: 20px;'>End-to-End ASR, Text Generation, and TTS</p>")
+    gr.HTML("<center>Upload or record audio. The model will transcribe, generate a response, and read it out.</center>")
+
+    audio_input = gr.Audio(label="Input Audio", type="filepath")
+    text_output = gr.Textbox(label="Transcription")
+    generated_text_output = gr.Textbox(label="Generated Text")
+    audio_output = gr.Audio(label="Output Speech")
+    submit_btn = gr.Button("Submit")
+
+    submit_btn.click(
+        fn=process_audio,
+        inputs=audio_input,
+        outputs=[text_output, generated_text_output, audio_output]
+    )
+
+if __name__ == "__main__":
+    demo.launch()