File size: 3,696 Bytes
5b74a4b 1ee8cb6 83e3ccb abd2b24 ea3653e 5b74a4b 25fb027 47c691b 25fb027 a927d1d ea3653e 1ee8cb6 ea3653e 5add931 aef1f22 ea3653e 1ee8cb6 776c3e1 47453aa 25fb027 1ee8cb6 952235c 25fb027 393002d 25fb027 393002d 25fb027 393002d 5b74a4b 5add931 72632b9 25fb027 5add931 25fb027 c58bd88 8c23bfa 5add931 25fb027 17cfe18 25fb027 a5ec736 b2c7d3a 5b74a4b ea3653e 8fe6fd5 5b74a4b b2c7d3a |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 |
import gradio as gr
from transformers import pipeline, AutoTokenizer
import numpy as np
from pydub import AudioSegment
import librosa
# Load the pipeline for speech recognition and translation
pipe = pipeline(
"automatic-speech-recognition",
model="Akashpb13/Hausa_xlsr",
tokenizer="Akashpb13/Hausa_xlsr"
)
translator = pipeline("text2text-generation", model="Baghdad99/saad-hausa-text-to-english-text")
tts = pipeline("text-to-speech", model="Baghdad99/english_voice_tts")
def translate_speech(audio_input):
print(f"Type of audio: {type(audio_data_tuple)}, Value of audio: {audio_data_tuple}") # Debug line
# Check if the input is a tuple (recorded audio) or a string (uploaded file)
if isinstance(audio_input, tuple):
# Extract the audio data from the tuple
sample_rate, audio_data = audio_input
else:
# Load the audio file as a floating point time series
audio_data, sample_rate = librosa.load(audio_input, sr=None)
# Normalize the audio data to the range [-1, 1]
audio_data_normalized = audio_data / np.iinfo(audio_data.dtype).max
# Convert the normalized audio data to float64
audio_data_float64 = audio_data_normalized.astype(np.float64)
# Use the speech recognition pipeline to transcribe the audio
output = pipe(audio_data_float64)
print(f"Output: {output}") # Print the output to see what it contains
# Check if the output contains 'text'
if 'text' in output:
transcription = output["text"]
else:
print("The output does not contain 'text'")
return
# Print the transcription
print(f"Transcription: {transcription}")
# Use the translation pipeline to translate the 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
# Print the translated text string
print(f"Translated text string: {translated_text_str}")
# 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
# Flatten the audio data
synthesised_speech_data = synthesised_speech_data.flatten()
# Print the shape and type of the synthesised speech data
print(f"Synthesised speech data type: {type(synthesised_speech_data)}, Synthesised speech data shape: {synthesised_speech_data.shape}")
# Scale the audio data to the range of int16 format
synthesised_speech = (synthesised_speech_data * 32767).astype(np.int16)
return 16000, synthesised_speech
# Define the Gradio interface
iface = gr.Interface(
fn=translate_speech,
inputs=gr.inputs.Audio(source="microphone", type="file"), # Change this line
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."
)
iface.launch()
|