Our expertise, in a few words
Proprietary speech recognition engine built on spontaneous conversations
100% proprietary transcription, more accurate that the rest of the market
Confirmed by external benchmarks
Automatic adjustment of sentences in real time thanks to dedicated language models
Combination of Machine/Deep Learning and human analysis of conversations
Very large library of vocabulary adapted to any sectors/industries
Science, in the heart of our solution
Manual moderation or corpus handling
- Confrontation and transparency on the Word Error Rate through evaluation corpus
- Realization of unique training and evaluation corpuses per client
- Multiple speakers and conversational analysis processing (intonation, flow, emotion, hesistancy…)
Segmentation or signal transcription
- Dedicated and specialized language models
- Learning phrasing specific to a brand/industry
- Improving an acoustic model continuously by learning based on accents and ambient noise
Transparency of our Word Error Rate
- Scientific method of standard calculation (Word Error Rate)
- Confidence scores for each transcribed conversation
- A unique Word Error Rate for each adapted model
Science to the business applications
- PhDs specialized in Automatic Speech Recognition (ASR)
- Dedicated annotation department: linguists, data builders and data scientists
- Being accompanied by a professional holding his PhD in acoustics and emotion processing
- Contact-center and marketing expert teams
- Telecoms and VoIP engineers
Our achievements
"Better-Than-Human" Voice recognition
- Pole position in France: four times better than the market Word Error Rate (WER)
- Average word error rate (WER): 4.1%, better than human (5% to 7.5%)
- Retranscribing Speech-to-Text in real time: 0.72-second lag enables usage of a voicebot
Telecom operator
- Native speaker diarisation through the recorder
- 0% of voice data lost (vs. 13% average)
- We could plug in directly for an even lower lag
Roadmap
- More than 35 features in development (number of pauses, response delays, etc.)
- Inclusion of new languages (Portuguese, German etc.) in progress
- End-to-end contextual call classification—signal to label
Recognized expertise
GDPR, Health Data & other personnaly identifiable info. (PII)
4 technical patents filed in the US in 2019
Technical benchmarks by
Orange Labs, WebHelp, Groupama, AXA
A couple dozen open-source repositories (GitHub)
Certified French operator (ARCEP—Comparable to Federal Communications Commission (FCC))
Integration with all your tools & telephony
Connection methods to audio flow & data
- API Connect & Activate available to you
- Native integration through a simple tag
- Plug-&-play connectors to the following:
- Analytics platforms (Adobe Analytics)
- Existing Ad Servers (Google Ads, Facebook Ads, etc.)
Phone-connection methods
- Call tracking (SDA)
- Simple call routing
- Session Recording Protocol (SIPREC)
- SIP TRUNK