dipsankar-sinha/Wav2Vec2-ASR-Amharic-CV-16.1

This is the code for fine tuning Wav2Vec2-Large-XLS-R-300M for building an ASR for Amharic Language using Mozilla Common Voice 16.1 Dataset. Implementation of the model to a web application using Gradio Library. # Wav2Vec2-ASR-Amharic-CV-16.1 This is the code for fine tuning Wav2Vec2-Large-XLS-R-300M for building an ASR for Amharic Language using Mozilla Common Voice 16.1 Dataset. Implementation of the model to a web application using Gradio Library. - license: apache-2.0 - Base Model: facebook/wav2vec2-xls-r-300m - Datasets: common_voice_16_1 This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice_16_1 dataset. It achieves the following results on the evaluation set: - Validation Loss : 1.6333 - Word Error Rate (WER): 0.8639 ## Model description # Wav2Vec2-XLS-R-300M Facebook's Wav2Vec2 XLS-R counting **300 million** parameters. XLS-R is Facebook AI's large-scale multilingual pretrained model for speech (the "XLM-R for Speech"). It is pretrained on 436k hours of unlabeled speech, including VoxPopuli, MLS, CommonVoice, BABEL, and VoxLingua107. It uses the wav2vec 2.0 objective, in 128 languages. When using the model make sure that your speech input is sampled at 16kHz. **Note**: This model should be fine-tuned on a downstream task, like Automatic Speech Recognition, Translation, or Classification. Check out **this blog** for more information about ASR. XLS-R Paper Authors: Arun Babu, Changhan Wang, Andros Tjandra, Kushal Lakhotia, Qiantong Xu, Naman Goyal, Kritika Singh, Patrick von Platen, Yatharth Saraf, Juan Pino, Alexei Baevski, Alexis Conneau, Michael Auli ## Intended uses & limitations - This model is a general purpose speech to text conversion model. There can be several use cases of this model such as in eduation, business, healthcare, government sectors etc.The model can be specifically designed for these sectors. - The primary limitation is the WER is not as per with the standard. That is because of lack of data. The model's performance can be improved by using a large dataset with complex sentences and sector specific Information. ## Training and evaluation data - Datasets: common_voice_16_1 - Language Used: Amharic …