AmazingMrBill/ASR-swahili
Domain:
natural language processingRecord type:
model
Creator:
Ama
Host:
# Swahili Automatic Speech Recognition (ASR) Model
## Overview
This repository contains code and instructions for utilizing a Swahili Automatic Speech Recognition (ASR) model based on the transformers library. The model is pre-trained on Swahili audio data and can be used to transcribe Swahili speech in various applications.
## Instructions
### 1. Install Dependencies
Make sure you have the required dependencies installed. You can install them by running the following command:
```bash
!pip install transformers
```
### 2. Set Up the ASR Pipeline
```python
from transformers import pipeline
from google.colab import drive
# Set up the ASR pipeline using the Swahili model
# model example is Akash/Swahili_xlrs from hugging face, I also used alamsher/wav2vec2-large-xlsr-53-common-voice-sw
pipe = pipeline("automatic-speech-recognition", model="Akashpb13/Swahili_xlsr", device=0)
# Mount Google Drive to access your files
drive.mount('/content/drive')
```
### 3. Load and Transcribe Audio Files
#### Single Audio File:
```python
from IPython.display import Audio
# Display an audio file
Audio("/content/test/common_voice_sw_27729935.mp3")
# Transcribe the audio file
pipe("/content/test/common_voice_sw_27729935.mp3")
```
#### Multiple Audio Files:
```python
# Create a DataFrame with file paths
test = pd.read_csv('/content/drive/MyDrive/Models/SampleSubmission.csv')
test["my_path"] = ["/content/test/" + i for i in test.path]
# Transcribe multiple audio files
pipe(test.my_path.to_list())
```
### 4. Save Transcriptions to CSV
```python
# Create a submission DataFrame
sub = pd.DataFrame()
sub["path"] = test.path.to_list()
sub["sentence"] = result_list
# Save the submission to a CSV file
sub.to_csv("ASR_Submission.csv", index=False)
```
### 5. Perform ASR on Entire Dataset
```python
# Perform ASR predictions on the entire dataset
res = []
for path in tqdm(test.path):
res.append(pipe(f'/content/drive/MyDrive/asr/test_audios/{path}')['text'])
# Update the test Data …