When choosing an AI phone agent or voicebot, the natural question is “what is its recognition accuracy?” But the figure changes with what was measured and on what audio, and not many products publish one. This article sets out what major products in Japan publish, and voicast’s results.
Vendor information in this article is based on each company’s official website as of 17 October 2026. Please check the official websites for the latest information. voicast is a service operated by SHANNON.
voicast’s recognition accuracy | Standard Japanese 94%, dialects 93%
The recognition accuracy of voicast, our voice AI agent platform, in our internal tests:
| Recognition accuracy | |
|---|---|
| Standard Japanese | 94% |
| Dialects | 93% |
*Values from our internal tests measuring how accurately numbers and names are recognized. We measured under several conditions and show the highest value (254 patterns, 1,016 test runs; the values shown were measured at phone audio quality; dialects were measured on public Japanese speech data that includes dialects).
What AI phone and voicebot vendors publish | What each figure measures
We checked whether major AI phone, voicebot and speech recognition products in Japan publish recognition accuracy on their official websites. This is not a comparison of which is more accurate.
| Product (company) | Accuracy published | What the figure measures |
|---|---|---|
| IVRy (株式会社IVRy) | Not found on the official website (as of 17 October 2026) | – |
| AI Messenger Voicebot / AI Worker VoiceAgent (株式会社AI Shift) | Published for some items | Address recognition accuracy; recognition rate for order information containing letters |
| PKSHA VoiceAgent (株式会社PKSHA Technology) | Not found on the official website (as of 17 October 2026) | – |
| AmiVoice (株式会社アドバンスト・メディア) | Published (company tech blog) | Character-level recognition accuracy |
| MiiTel (株式会社RevComm) | Published (company’s own study) | Proportion of characters with errors |
| commubo (株式会社ソフトフロントジャパン) | Improvement published | Improvement in the error rate for personal names, addresses and short sentences compared with the previous engine (company comparison) |
| voicast (SHANNON) | Published | Recognition of numbers and names |
All sources are the companies’ official websites or official announcements, checked on 17 October 2026 (URLs are listed under “Sources” at the end).
Published figures differ in what they measure and under what conditions, so they cannot be compared side by side as they are. When you look at a figure, check what it measures, and in the end, try the product on your own calls.
Summary
Some AI phone and voicebot products publish recognition accuracy and some do not, and what is published varies. voicast publishes Standard Japanese 94% and dialects 93%, together with what was measured and how many test runs were made.
SHANNON’s voicast is a voice AI agent platform: AI answers phone, web and app audio and returns the captured items and a summary by API and webhook. If you are considering it, please contact us.
Sources
- IVRy official website (checked 17 October 2026)
- AI Shift: announcement of the address recognition feature for AI Messenger Voicebot (PR TIMES, 5 January 2022, Japanese) (checked 17 October 2026)
- AI Shift: Linbell case study for AI Worker VoiceAgent (PR TIMES, 2 September 2026, Japanese) (checked 17 October 2026)
- PKSHA VoiceAgent official website (checked 17 October 2026)
- Advanced Media, AmiVoice Tech Blog: recognition accuracy on call-centre audio (Japanese) (checked 17 October 2026)
- Advanced Media, AmiVoice Tech Blog: recognition accuracy on meeting audio (Japanese) (checked 17 October 2026)
- RevComm: announcement of its own speech recognition study for MiiTel (18 January 2024, Japanese) (checked 17 October 2026)
- SoftFront Holdings: announcement of the new speech recognition engine for commubo (20 May 2024, PDF, Japanese) (checked 17 October 2026)