Lance VOX – Speech Recognition Evolved 🎤
🚀 Lance VOX is a custom-built speech recognition model, designed to transcribe spoken audio into text efficiently. Lance VOX is optimized for local and cloud inference with transformers and PyTorch.
🌟 Key Features
- ✅ Custom transformer-based ASR architecture
- ✅ Supports Hugging Face’s transformers library
- ✅ Lightweight and optimized for fast transcription
- ✅ Fully trainable on custom datasets
- ✅ Designed to evolve with larger datasets and enhanced features
📥 Installation & Setup
You can load Lance VOX using transformers:
from transformers import AutoTokenizer, AutoFeatureExtractor, AutoModelForSeq2SeqLM import librosa
model_name = "NeuraCraft/LanceVox" tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True) feature_extractor = AutoFeatureExtractor.from_pretrained(model_name, trust_remote_code=True) model = AutoModelForSeq2SeqLM.from_pretrained(model_name, trust_remote_code=True)
audio_path = "your_audio_file.wav" speech_array, sr = librosa.load(audio_path, sr=16000) input_features = feature_extractor(speech_array, sampling_rate=sr, return_tensors="pt").input_features
generated_ids = model.generate(input_features) transcription = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0] print(transcription)
🛠 How to Use Lance VOX
1️⃣ Direct Audio Transcription
audio_path = "sample.wav" speech_array, sr = librosa.load(audio_path, sr=16000) inputs = feature_extractor(speech_array, sampling_rate=sr, return_tensors="pt").input_features output_ids = model.generate(inputs) transcription = tokenizer.batch_decode(output_ids, skip_special_tokens=True)[0] print(transcription)
2️⃣ Fine-tuning for Custom Datasets
from transformers import Seq2SeqTrainer, Seq2SeqTrainingArguments
training_args = Seq2SeqTrainingArguments( output_dir="./lance_vox_finetuned", per_device_train_batch_size=8, per_device_eval_batch_size=8, num_train_epochs=3, save_steps=500, )
trainer = Seq2SeqTrainer( model=model, args=training_args, train_dataset=your_dataset, eval_dataset=your_eval_dataset, tokenizer=tokenizer, )
trainer.train()
📊 Performance & Evaluation
Lance VOX is in active development. Early evaluations focus on:
🔹 Word Error Rate (WER) – Measures transcription accuracy
🔹 Real-time transcription latency
🔹 Token prediction accuracy from audio
✅ Planned Enhancements
🔹 Support for multi-language transcription
🔹 Integration with real-time audio streams
🔹 Larger, more diverse training datasets
🔹 Multimodal input with text & speech
🚀 Future Roadmap
Lance VOX is designed to evolve alongside Lance AI:
Planned Features:
🔜 Real-time transcription with minimal latency
🔜 Continuous learning from user corrections
🔜 Multilingual support
🔜 Advanced integration with Lance AI for speech-to-text and command execution
🏗 Development & Contributions
Lance VOX is developed by NeuraCraft. Contributions, dataset suggestions, and feedback are welcome!
Contact & Updates: Developer: NeuraCraft Project Status: 🚧 In Development