Faster Speech To Speech:

[]

A very fast way to get S2S models

How? just abuse APIs (while they are still free)

Important: Currently it is still in development, overtime I would increase its feature capabilities and modifications :) <If my school hours permits ?>

A super fast and quick way to implement a speech to speech model using:

1. Groq

2. Google TTS Service

If you wish to just call Google TTS API just run

python s2s.py --gtts

3. Coqui TTS (Locally Run & Modifiable!) [Default]

Simple Pipeline:

Person Talking ===> Groq Whisper ===> Groq Llama-70b-8192 ===> Google TTS Services

Modifications:

How To Use:

1. Ensure you install the requirements via

   pip install requirements.txt

2. Remember to sign up for Groq Playground and create an API_KEY

3. Run the script

   python s2s.py

It goes by a simple CLI interface

Just run python s2s.py followed by the following args commands you can try:

--output file_to_path : To set path for audio

--gtts: Enables Google TTS Service, else fallsback to Coqui TTS

--temperature (insert int): controls creativity of the model

--audio (audio_name): Sets name of your microphone

--model (model_name): Set model name (Default is llama3-70b-8192)

Requirements:

1. Cuda 12.4

2. Groq Account

3. Accepting Coqui TTS UAT

Acknowledgements:

1. Groq (For providing superfast inference)

2. Coqui (For providing the open source weights)