Update app.py

app.py

@@ -1,68 +1,64 @@
 import gradio as gr
-import requests
+from transformers import pipeline, AutoTokenizer
 import numpy as np
-from pydub import AudioSegment
-import io
-from IPython.display import Audio
 
-# Define the Hugging Face Inference API URLs and headers
-ASR_API_URL = "https://api-inference.huggingface.co/models/Baghdad99/saad-speech-recognition-hausa-audio-to-text"
-TTS_API_URL = "https://api-inference.huggingface.co/models/Baghdad99/english_voice_tts"
-TRANSLATION_API_URL = "https://api-inference.huggingface.co/models/Baghdad99/saad-hausa-text-to-english-text"
-headers = {"Authorization": "Bearer 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"}
-
-# Define the function to query the Hugging Face Inference API
-def query(api_url, payload=None, data=None):
-    if data is not None:
-        response = requests.post(api_url, headers=headers, data=data)
-    else:
-        response = requests.post(api_url, headers=headers, json=payload)
-    response_json = response.json()
-    if 'error' in response_json:
-        print(f"Error in query function: {response_json['error']}")
-        return None
-    return response_json
+# Load the pipeline for speech recognition and translation
+pipe = pipeline(
+    "automatic-speech-recognition",
+    model="Baghdad99/saad-speech-recognition-hausa-audio-to-text",
+    tokenizer="Baghdad99/saad-speech-recognition-hausa-audio-to-text"
+)
+translator = pipeline("text2text-generation", model="Baghdad99/saad-hausa-text-to-english-text")
+tts = pipeline("text-to-speech", model="Baghdad99/english_voice_tts")
 
 # Define the function to translate speech
-def translate_speech(audio_file):
-    print(f"Type of audio: {type(audio_file)}, Value of audio: {audio_file}")  # Debug line
+def translate_speech(audio):
+    # Separate the sample rate and the audio data
+    sample_rate, audio_data = audio
 
-    # Use the ASR pipeline to transcribe the audio
-    data = audio_file.read()
-    output = query(ASR_API_URL, data=data)
-    print(f"Output: {output}")  # Debug line
+    # Use the speech recognition pipeline to transcribe the audio
+    output = pipe(audio_data)
+    print(f"Output: {output}")  # Print the output to see what it contains
 
-    # Check if output is not None
-    if output is not None:
-        # Check if 'error' key exists in the output
-        if 'error' in output:
-            print(f"Error: {output['error']}")
-            return
-
-        # Check if 'text' key exists in the output
-        if 'text' in output:
-            transcription = output["text"]
-        else:
-            print("Key 'text' does not exist in the output.")
-            return
+    # Check if the output contains 'text'
+    if 'text' in output:
+        transcription = output["text"]
     else:
-        print("Output is None.")
+        print("The output does not contain 'text'")
         return
 
     # Use the translation pipeline to translate the transcription
-    translated_text = query(TRANSLATION_API_URL, {"inputs": transcription})
+    translated_text = translator(transcription, return_tensors="pt")
+    print(f"Translated text: {translated_text}")  # Print the translated text to see what it contains
+
+    # Check if the translated text contains 'generated_token_ids'
+    if 'generated_token_ids' in translated_text[0]:
+        # Decode the tokens into text
+        translated_text_str = translator.tokenizer.decode(translated_text[0]['generated_token_ids'])
+    else:
+        print("The translated text does not contain 'generated_token_ids'")
+        return
+
+    # Use the text-to-speech pipeline to synthesize the translated text
+    synthesised_speech = tts(translated_text_str)
+    print(f"Synthesised speech: {synthesised_speech}")  # Print the synthesised speech to see what it contains
+
+    # Check if the synthesised speech contains 'audio'
+    if 'audio' in synthesised_speech:
+        synthesised_speech_data = synthesised_speech['audio']
+    else:
+        print("The synthesised speech does not contain 'audio'")
+        return
 
-    # Use the TTS pipeline to synthesize the translated text
-    response = requests.post(TTS_API_URL, headers=headers, json={"inputs": translated_text})
-    audio_bytes = response.content
+    # Scale the audio data to the range of int16 format
+    synthesised_speech = (synthesised_speech_data * 32767).astype(np.int16)
 
-    # Display the audio output
-    return Audio(audio_bytes)
+    return 16000, synthesised_speech
 
 # Define the Gradio interface
 iface = gr.Interface(
     fn=translate_speech, 
-    inputs=gr.inputs.File(type="file"),  # Change this line
+    inputs=gr.inputs.Audio(source="microphone", type="numpy"), 
     outputs=gr.outputs.Audio(type="numpy"),
     title="Hausa to English Translation",
     description="Realtime demo for Hausa to English translation using speech recognition and text-to-speech synthesis."