In the realm of speech-to-text technology, the accuracy and precision of transcription services play a pivotal role. At CCflow.ai we take pride in our commitment to excellence and continuous improvement. In this blog post, we shed light on our groundbreaking model by comparing it with competitors in the industry. From dataset size and license plate codes to the nitty-gritty of transcription accuracy, we delve into the nuances that set us apart.
Our model has been rigorously trained on a dataset comprising 227 samples, meticulously curated to ensure diversity and relevance. This robust dataset provides a solid base for comparison with competitors against our cutting-edge speech recognition systems.
The sample dataset contains license plate codes, specifically those from telephone calls in the Czech Republic. The codes consist of 7 characters, mixture of both alphabetical and numeric characters. The dataset allows us to demonstrate the nuances of real-world scenarios, making our model excellent at handling diverse and complex data.
To conduct a comprehensive evaluation, we integrated competitors’ APIs using the speech_recognition Python package. This enabled a direct comparison with our competitors and showcased our strengths.
One of the key differentiators is our proprietary DomainFormat function. When discrepancies arise in transcription results, our function kicks in to rectify them. It efficiently removes empty spaces, converts words and numbers to the correct format, and ensures uniformity across all results. Moreover, we’ve applied the function to our competitors’ models, to help them achieve the same results of the expected format.
To provide a holistic perspective, we tested our CCFlow STT model against prominent competitors, including Microsoft Azure Speech, Google Speech Recognition, and Google Cloud Speech API. The results speak volumes about the superiority of our model, validated through rigorous testing across multiple platforms.
Calculated by comparing the stt transcriptions after DomainFormat with the transcriptions fixed by human
- CCflow STT model: 80.62 %
- Microsoft Azure Speech: 56.83 %
- Google Speech Recognition: 44.49 %
- Google Cloud Speech API: 34.36 %
Source results from our script:
Evaluation of competitors finished, results saved locally
azure results:
{'total_count': 227, 'first_shot_accuracy': 0.5682819383259912, 'alternatives_accuracy': 0.5682819383259912, 'evaluation_result': {'FAILED': {'count': 98, 'percentage': 0.43171806167400884}, 'FATAL_ERROR': {'count': 0, 'percentage': 0.0}, 'ALTERNATIVES': {'count': 0, 'percentage': 0.0}, 'OK': {'count': 129, 'percentage': 0.5682819383259912}}}
google results:
{'total_count': 227, 'first_shot_accuracy': 0.44493392070484583, 'alternatives_accuracy': 0.44493392070484583, 'evaluation_result': {'FAILED': {'count': 126, 'percentage': 0.5550660792951542}, 'FATAL_ERROR': {'count': 0, 'percentage': 0.0}, 'ALTERNATIVES': {'count': 0, 'percentage': 0.0}, 'OK': {'count': 101, 'percentage': 0.44493392070484583}}}
google_cloud results:
{'total_count': 227, 'first_shot_accuracy': 0.3436123348017621, 'alternatives_accuracy': 0.3436123348017621, 'evaluation_result': {'FAILED': {'count': 149, 'percentage': 0.6563876651982379}, 'FATAL_ERROR': {'count': 0, 'percentage': 0.0}, 'ALTERNATIVES': {'count': 0, 'percentage': 0.0}, 'OK': {'count': 78, 'percentage': 0.3436123348017621}}}
Evaluation of competitors finished, results saved locally
ccflow results:
{'total_count': 227, 'first_shot_accuracy': 0.8061674008810573, 'alternatives_accuracy': 0.8061674008810573, 'evaluation_result': {'FAILED': {'count': 44, 'percentage': 0.19383259911894274}, 'FATAL_ERROR': {'count': 0, 'percentage': 0.0}, 'ALTERNATIVES': {'count': 0, 'percentage': 0.0}, 'OK': {'count': 183, 'percentage': 0.8061674008810573}}} In the fast-evolving landscape of speech-to-text technology, CCflow.ai stands as a beacon of innovation and reliability. Our commitment to excellence is not just a claim but a demonstrated reality through meticulous testing, groundbreaking functionality, and collaborative efforts with competitors. As we continue to push the boundaries of what’s possible, our copyrighted insights pave the way for a future where precision and accuracy define the standard in speech recognition technology. Contact us for a free trial and save operational costs for your company.