Help with training own language model using DeepSpeech
rajib-raiyat opened this issue · comments
I am interested in training my own Bangla language model using DeepSpeech, but I'm not sure where to start. I have my own Bangla dataset with audio and text, and I would like to use it to train a model that can transcribe Bangla speech to text offline. I am looking for guidance on how to preprocess my data, train a model, and evaluate its performance.
Can someone please provide detailed instructions or point me to a tutorial or guide that can help me with this process? Here are some specific questions I have:
- What are the best practices for preprocessing Bangla audio and text data for use with DeepSpeech?
- How do I create a Bangla language model and generate the necessary files for training a DeepSpeech model?
- What are the recommended training parameters and settings for training a Bangla language model using DeepSpeech?
- How do I evaluate the performance of my trained model, and what metrics should I use?
- Installation guidelines from scratch.
I would appreciate any help or advice that can be provided. Thank you in advance!
Hey Komol Kunty Rajib, you can do this by making the training model consume the Bangla dataset with audio and tex mapping to the model. Now create a superset model where it takes audio to text API to convert and take the audio and make clear text input to model and since model has trained on your data set it will provide the value mapping of the provided text key and easily retrieve the translation text.
@rajib-raiyat The DeepSpeech Playbook has been produced to assist you with training a model on other languages.