Python Project: Audio Transcription and Text-to-Speech Conversion Using Wav2Vec2 and Pyttsx3

at

 Explore an advanced Python project that combines audio transcription and text-to-speech synthesis using state-of-the-art tools like Librosa, PyTorch, and Hugging Face's Transformers library. This script demonstrates how to load and resample audio files, transcribe speech to text using Facebook's Wav2Vec2 model, and convert text back to speech with customizable voice options using pyttsx3. Perfect for anyone interested in speech processing, AI-driven voice technology, or natural language processing projects. Ideal for enhancing your Python skills and diving into real-world applications of AI in audio analysis.



import librosa

from scipy.signal import resample

import torch

from transformers import Wav2Vec2ForCTC, Wav2Vec2Tokenizer

import pyttsx3

from scipy.signal import resample


# Load audio file

audio_file = "directory of audio file"

audio, sr = librosa.load(audio_file, sr=None)



def resample_audio(audio, orig_sr, target_sr):

    duration = audio.shape[0] / orig_sr

    target_length = int(duration * target_sr)

    resampled_audio = resample(audio, target_length)

    return resampled_audio


# Example usage:

# resampled_audio = resample_audio(audio, 48000, 16000)


# Resample if necessary

if sr != 16000:

    audio = resample_audio(audio, sr, 16000)

    sr = 16000


print("Audio loaded and resampled successfully.")


# Load Wav2Vec2 model and tokenizer

tokenizer = Wav2Vec2Tokenizer.from_pretrained("facebook/wav2vec2-base-960h")

model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-base-960h")


print("Model loaded successfully.")


# Tokenize input

input_values = tokenizer(audio, return_tensors="pt").input_values


# Perform inference

with torch.no_grad():

    logits = model(input_values).logits


# Get predicted ids

predicted_ids = torch.argmax(logits, dim=-1)


# Decode the ids to text

transcription = tokenizer.batch_decode(predicted_ids)[0]

print("Transcription: ", transcription)


# Text-to-Speech

def text_to_speech(text, voice_gender='female', rate=150):

    engine = pyttsx3.init()

    voices = engine.getProperty('voices')

    

    if voice_gender == 'male':

        engine.setProperty('voice', voices[0].id)

    else:

        engine.setProperty('voice', voices[1].id)

    

    engine.setProperty('rate', rate)

    engine.say(text)

    engine.runAndWait()


# Example usage

long_text = "hi what happened"

text_to_speech(long_text, voice_gender='male', rate=150)  # For male voice

text_to_speech(long_text, voice_gender='female', rate=180)  # For female voice


print("Text-to-Speech conversion completed.")



#PythonProject
#AudioTranscription
#TextToSpeech
#Wav2Vec2
#PyTorch
#Librosa
#NLP
#SpeechRecognition
#VoiceSynthesis
#AIinPython
#NaturalLanguageProcessing
#SpeechToText
#Pyttsx3
#MachineLearning
#DeepLearning
#AudioProcessing
#PythonAI
#TransformersLibrary
#PythonCoding
#PythonTutorial

Labels:

Featured Post

Day 41 — BGP Confederations: Sub-AS Design, External View and Migration

1. Opening Confederations are another way to scale BGP inside a large administrative domain. They divide the domain into member autonomous systems while presenting a single confederation identifier to external peers. They are powerful, but their operational model is more complex than simply 'using private ASNs inside.' The engineering goal is not to memorize another BGP command. It is to understand what information each speaker is allowed to propagate, what path information can be hidden, and what failure domain is created by the chosen control-plane architecture . 2. Concept and standards behavior RFC 5065 defines AS_CONFED_SEQUENCE and AS_CONFED_SET and how member-AS relationships are represented. Confederation external sessions have eBGP-like properties inside the confederation, while the confederation is presented externally as one AS. Modern guidance must also account for the fact that RFC 9774 prohibits new origination of AS_SET/AS_CONFED_SET in ordinary aggregation c...