objective her is to develop a machine learning ASR model capable of accurately recognizing spoken words in Swahili.
# ASR_Notebook.
The Objective here is to develop a machine learning ASR model capable of accurately recognizing spoken words in Swahili.
# Dataset.
I have 4 files.
- test.csv - This is the dataset that will be used for model testing.
- train.csv - Train contains the target. This is the dataset that that will be used to train the model.
- test_audios.zip - Test audios that will be used to test the model.
- train_audios.zip - Train the audios that will be used to train the model.
### Step 1:
To start with the problem , I look on data. CSV files and Audio Data. I converted mp3 audio data to wav file format. I thought this being compactible and best format with Machine Learning Algorithm for this.
### Step 2:
After uploading csv files we convert them into intergers , then encoding and decoding functions were defined , data were splitted into training and validation.
### Step 3:
This step we define CTC Loss function , model function after that decoding of test data were defined.
### Step 4:
Model training and after that we check the result using our validation data.
# Progress
With the notebook given here in the table, I have built the pipeline. But when I finally trained the model, I got no predictions.