Announcing New v2 ASR Model in Temi
Temi has always been about making high-quality automated transcription available to everyone, and speech recognition accuracy is at the heart of this goal. Today, we’re happy to announce that Temi has fully transitioned to Rev’s v2 ASR model, which offers an over 25% increase in accuracy when compared to our previous model.
Rev’s v2 model applies the latest deep learning techniques to our millions of hours of transcribed speech, ingesting audio directly and learning as it goes. This approach is the result of two years of technical research and development by our speech team, and it largely solves key problems in accuracy, training, pronunciation, accents and diarization.
With Rev’s v2 ASR model, Temi is now significantly better at handling:
- different pronunciations
- different acoustic environments
- multi-speaker audio
- regional accents/dialects
- non-native speech
This means that, regardless of whether you’re conducting an interview, recording a focus group discussion or sending yourself voice notes, you’ll find that your transcripts are returned faster and with fewer errors related to pronunciation, accents and speaker separation.
To illustrate, here are a couple of real-world transcription examples showing these improvements in practice:
Earlier: You’re going to avoid this in your code by [inaudible] easily recognized that the fires
Now: You can avoid this in your own code by preferring easily recognized identifiers
Earlier: despite the impact, let me assure you that our bank remains strong and focused first on capital and really level some solid and well above the minimum dilatory levels
Now: Despite the impact. Let me assure you that our bank remains strong and focused. First, our capital and liquid levels are solid and well above the minimum regulatory levels
So Temi is faster and better…but by how much?
- We improved transcription quality by 26% overall.
- For audio with frequent speaker switches, our quality is even higher (35% improvement).
- Transcription is faster by 33-40%.
Using v2 ASR in Temi
Wondering how you can get all these benefits for your next Temi transcription order? Short answer: you already have them!
For the last two months, we’ve been progressively switching Temi customers to the new v2 model and as of today, all Temi customers are using the v2 model by default. So, all the benefits listed above are already available and active on your Temi account, no action needed.
We’re always working to improve Temi for our users, and we expect our switch to Rev’s v2 ASR model to result in quicker, more accurate results and a better overall experience. We’d love to hear your feedback – let us know by emailing us at email@example.com.