NuralVoiceSTT

NuralVoiceSTT

Developed by Blink Digital

Accurate universal English speech-to-text model optimized for both callcenter and wideband audio scenarios.

Model Details

NuralVoiceSTT is a state-of-the-art English (US) speech recognition model developed by Blink Digital. This model delivers high-accuracy speech-to-text conversion with support for both narrowband (callcenter) and wideband audio formats.

Key Features

• High Accuracy: Optimized for real-world speech recognition tasks
• Universal Support: Works with both callcenter and wideband audio
• Dynamic Graph: Advanced dynamic graph architecture for flexible recognition
• Speaker Adaptation: Built-in i-vector support for speaker adaptation

Model Architecture

• Acoustic Model: Advanced neural network-based acoustic modeling
• Language Model: Large vocabulary language model with dynamic graph
• Feature Extraction: MFCC-based feature extraction pipeline
• Speaker Adaptation: I-vector based speaker adaptation system

Model Structure

• am/ - Acoustic model files
• conf/ - Configuration files (MFCC, model config)
• graph/ - Language model graph files (FST files, phones, words)
• ivector/ - I-vector files for speaker adaptation

Usage

Python API

import json
import wave
from vosk import Model, KaldiRecognizer, SetLogLevel

# Set log level
SetLogLevel(0)

# Load the NuralVoiceSTT model
model = Model("path/to/NuralVoiceSTT")

# Process audio file
wf = wave.open("audio.wav", "rb")
rec = KaldiRecognizer(model, wf.getframerate())
rec.SetWords(True)

# Recognize speech
while True:
    data = wf.readframes(4000)
    if len(data) == 0:
        break
    if rec.AcceptWaveform(data):
        result = json.loads(rec.Result())
        print(result["text"])

# Get final result
result = json.loads(rec.FinalResult())
print(result["text"])

Download from Hugging Face

from huggingface_hub import snapshot_download

model_path = snapshot_download(
    repo_id="ashishkblink/NuralVoiceSTT",
    local_dir="./models/NuralVoiceSTT"
)

Installation

pip install vosk

Performance

NuralVoiceSTT is optimized for:

• Callcenter Audio: High accuracy on narrowband telephone audio
• Wideband Audio: Excellent performance on high-quality audio recordings
• Real-time Processing: Efficient for live transcription applications

License

This model is distributed under the Apache License 2.0, which allows:

• ✅ Free use (including commercial)
• ✅ Modification
• ✅ Distribution

Citation

If you use this model, please cite:

@misc{nuralvoicestt2024,
  title={NuralVoiceSTT: High-Accuracy English Speech-to-Text Model},
  author={Blink Digital},
  year={2024},
  publisher={Blink Digital},
  url={https://huggingface.co/ashishkblink/NuralVoiceSTT}
}

About Blink Digital

NuralVoiceSTT is developed and maintained by Blink Digital, a leading provider of AI-powered speech recognition solutions. For more information, visit our Hugging Face profile.

Support

For questions, issues, or commercial inquiries, please contact Blink Digital through our Hugging Face profile.