Promote Sign in Sign up

Voice recognition

Conformer2 ✓

State-of-the-art speech recognition powered by 1.1M hours of data.

Source rating 5.0 (1) 1.2k views 957.2K/mo traffic 21 saves Freemium
Speech Recognition Audio Data Processing Latent Period Reduction
More categories (4)
Model Ensembling Alphanumeric Decoding Language Models Voice recognition
Visit website Sign in to save

Free + from Starting at $0.12/hr for Speech-to-Text

Verdict

A quick read before you dig into the full page.

Best if
State-of-the-art speech recognition powered by 1.1M hours of data
Skip if
Only trained on English
Pick this when
Trained on 1.1 million hours

Considering My Voice AI instead? Compare side by side

Overview

Conformer-2 is an advanced automatic speech recognition AI model developed as a successor to Conformer-1. It's designed with robust improvements for decoding proper nouns, alphanumerics, and exhibiting superior performance in noisy environments.

This has been achieved through intensive training on a large corpus of English audio data. An advantage of Conformer-2 is that it does not compromise on word error rate compared to Conformer-1, while providing enhanced user-oriented metrics.

Further improvements to Conformer-2, in comparison to its predecessor, were realized by augmenting the training data volume and increasing pseudo-label models.

Furthermore, with modifications to the inference pipeline, the latency period of Conformer-2 is reduced, thus expediting overall performance. Another critical step-up with Conformer-2 pertains to its innovative training technique that leverages model ensembling.

Instead of deriving labels solely from a single 'teacher', labels are generated in this model from multiple 'teachers', ensuring a more versatile and robust model.

This has the effect of reducing the impact of individual model failures. The development of Conformer-2 also involved an exploration into data and model parameter scaling, increasing the model size, and extending the training audio data.

These approaches were aimed at matching the underutilized potential identified by the 'Chinchilla' paper for large language models. With these updates, Conformer-2 provides faster response times than Conformer-1, bucking the trend of larger models being slower and more expensive.

Conformer2 website

Access & connections

Where you can use Conformer2.

Platforms

Web

Capabilities

What this tool accepts, produces, and is strongest at.

Input

  • API
  • Audio

Output

  • API
  • Text

Pros & cons

Pros

  • Trained on 1.1 million hours
  • Enhanced proper noun recognition
  • Improved alphanumeric recognition
  • Increased noise robustness
  • Utilizes model ensembling
  • Reduced processing times
  • Impressed user-oriented metrics
  • Ideal for speech-to-text transcriptions
Show 40 more
  • Significant model size enhancements
  • Large language model optimized
  • Reduced inference latency period
  • Excellence in handling individual model failures
  • Robust results on real-world data
  • Improved speed over predecessor
  • Optimized serving infrastructure
  • 31.7% alphanumeric improvement
  • 6.8% proper noun error rate improvement
  • 12.0% noise robustness improvement
  • Scaling up data and model parameters
  • Faster results delivery
  • Reduced variability
  • Improvements in transcribing numerical data
  • Enhanced noise handling abilities
  • Flexibility for continual experimentation
  • API parameters speech_threshold
  • Minimal API changes for users
  • Model can be tried in Playground
  • Optimized for most real use cases
  • Designed to reduce model's variance
  • Failure cases subdued by model ensembling
  • Enables faster overall performance
  • Delivers more readable transcripts
  • Large gains in Alphanumeric Transcription Accuracy
  • Shows reduced variance in character error rate
  • Improved performance in noisy environments
  • Training speed is 1.6x faster
  • Automatic rejection of low speech proportion files
  • Capable of handling wide distribution of data
  • Explores into multimodality and self-supervised learning
  • Integration with in-house hardware
  • Improved real-world applications
  • State-of-the-art speech recognition model
  • Reduced transcription time
  • Copes with robust noises
  • Capabilities in robustness improvement
  • Efficient model size scaling
  • Equipped for model/dataset scaling
  • Efficient model ensembling

Cons

  • Only trained on English
  • Potential bias from teachers
  • No multi-language support
  • Narrow training data focus
  • Dependent on ensembling technique
  • Problems with edge-case alphanumerics
  • May inconsistently handle noise
  • No small-scale application
Show 2 more
  • Requires substantial computational power
  • In-house infrastructure dependency

Compare alternatives

Similar tools in the catalog — with a clear reason to open each one.

No pricing

Full comparison

Why look Offers a public API

Full comparison

Free + from $22.50/mo

Full comparison

Free + from $12.00/mo

Full comparison

Free + from $8.00/mo

Full comparison

Free + from $9.99/mo

Full comparison

Releases 1

Initial release July 20, 2023 Current
Initial release of Conformer2.

FAQ 12

What is Conformer-2?
Conformer-2 is an advanced AI model designed for automatic speech recognition, developed as a successor to Conformer-1. It is particularly effective at recognizing proper nouns, alphanumerics, and is robust in noisy environments.
How is Conformer-2 different from its predecessor, Conformer-1?
Conformer-2 distinguishes itself from its predecessor, Conformer-1, through several key improvements. It has made significant advancements in decoding proper nouns, alphanumerics, and demonstrating superior performance in noisy situations. This has been achieved through extensive training on a vast quantity of English audio data. Moreover, it uses an enhanced training technique that employs model ensembling, generating labels from multiple strong 'teachers' instead of just one. This makes Conformer-2 more versatile and robust as it reduces the impact of individual model failures. Additionally, despite being a larger model, Conformer-2 offers faster response times compared to Conformer-1 due to optimizations in the inference pipeline.
What is the main function of Conformer-2?
The primary function of Conformer-2 is to provide automatic speech recognition. It transforms spoken data into text, making it an essential component for AI pipelines focusing on generative AI applications that use spoken data.
How much English audio data has Conformer-2 been trained on?
Conformer-2 has been trained on an extensive amount of 1.1 million hours of English audio data.
What enhancements does Conformer-2 provide in terms of speech recognition?
Conformer-2 offers enhanced recognition of proper nouns and alphanumerics. It also provides robustness to noise, thereby proving superior performance in real-world audio conditions which could be challenging.
What is model ensembling in the context of Conformer-2?
Model ensembling in the context of Conformer-2 is a training technique in which labels are generated from multiple strong 'teachers' rather than a single one, reducing variance and enhancing the model's performance when presented with unseen data during training.
How does Conformer-2's speed compare with that of Conformer-1?
Despite its increased model size, Conformer-2 offers a significant improvement in speed compared to Conformer-1. The serving infrastructure has been optimized to ensure faster processing times, achieving up to a 55% reduction in relative processing duration across all audio file durations.
What improvements does Conformer-2 offer in terms of user-oriented metrics?
Conformer-2 demonstrates significant enhancements in various user-oriented metrics. These include a 31.7% improvement on alphanumerics, a 6.8% improvement on the proper noun error rate, and a 12.0% improvement in noise robustness.
How does Conformer-2 perform in real-world applications?
In real-world applications, Conformer-2 shows immense capability. For instance, it achieves significantly lower error rates in recognizing proper nouns and alphanumeric data, which are often crucial in real-world use cases. Conformer-2 is also more robust to noise, making it well adapted to varied and potentially challenging audio conditions found in the real world.
What type of AI applications would benefit the most from Conformer-2?
AI applications focused on generative use of spoken data would benefit the most from Conformer-2. This model is ideal for generating accurate speech-to-text transcriptions, a crucial component for these types of AI applications.
Why does Conformer-2 use multiple 'teachers' for label generation?
Conformer-2 uses multiple 'teachers' for label generation to create a more robust and versatile model. This approach mitigates the influence of individual model failures, broadening the model's exposure to a wider distribution of behaviors.
How is the Conformer-2 training method innovative?
Conformer-2's training method is innovative because it uses model ensembling, generating labels from multiple teacher models instead of just one. This approach reduces variance and produces a model that is more robust when exposed to unseen data during training.
Conformer2 Free + from Starting at $0.12/hr for Speech-to-Text
Visit website