| HIYACCENT: An Improved Nigerian-Accented Speech Recognition System Based on Contrastive Learning | |
| The global objective of this research was to develop a more robust model for the Nigerian English Speakers whose English pronunciations are heavily affected by their mother tongue. For this, the Wav2Vec-HIYACCENT model was proposed which introduced a new layer to the Novel Facebook Wav2vec to capture the disparity between the baseline model and Nigerian English Speeches. A CTC loss was also inserted on top of the model which adds flexibility to the speech-text alignment. This resulted in over 20% improvement in the performance for NAE.T | |
| Fine-tuned facebook/wav2vec2-large on English using the UISpeech Corpus. When using this model, make sure that your speech input is sampled at 16kHz. | |
| The script used for training can be found here: https://github.com/amceejay/HIYACCENT-NE-Speech-Recognition-System | |
| ##Usage: The model can be used directly (without a language model) as follows... | |
| #Using the ASRecognition library: | |
| from asrecognition import ASREngine | |
| asr = ASREngine("fr", model_path="codeceejay/HIYACCENT_Wav2Vec2") | |
| audio_paths = ["/path/to/file.mp3", "/path/to/another_file.wav"] | |
| transcriptions = asr.transcribe(audio_paths) | |
| ##Writing your own inference speech: | |
| import torch | |
| import librosa | |
| from datasets import load_dataset | |
| from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor | |
| LANG_ID = "en" | |
| MODEL_ID = "codeceejay/HIYACCENT_Wav2Vec2" | |
| SAMPLES = 10 | |
| #You can use common_voice/timit or Nigerian Accented Speeches can also be found here: https://openslr.org/70/ | |
| test_dataset = load_dataset("common_voice", LANG_ID, split=f"test[:{SAMPLES}]") | |
| processor = Wav2Vec2Processor.from_pretrained(MODEL_ID) | |
| model = Wav2Vec2ForCTC.from_pretrained(MODEL_ID) | |
| # Preprocessing the datasets. | |
| # We need to read the audio files as arrays | |
| def speech_file_to_array_fn(batch): | |
| speech_array, sampling_rate = librosa.load(batch["path"], sr=16_000) | |
| batch["speech"] = speech_array | |
| batch["sentence"] = batch["sentence"].upper() | |
| return batch | |
| test_dataset = test_dataset.map(speech_file_to_array_fn) | |
| inputs = processor(test_dataset["speech"], sampling_rate=16_000, return_tensors="pt", padding=True) | |
| with torch.no_grad(): | |
| logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits | |
| predicted_ids = torch.argmax(logits, dim=-1) | |
| predicted_sentences = processor.batch_decode(predicted_ids) | |
| for i, predicted_sentence in enumerate(predicted_sentences): | |
| print("-" * 100) | |
| print("Reference:", test_dataset[i]["sentence"]) | |
| print("Prediction:", predicted_sentence) | |