Skip to main content

Speech Recognition & AI

Bilingual Speech-to-Text (Preview)

We’ve developed real-time automatic language detection and code-switching between Japanese and English. VoicePing can now seamlessly transcribe multilingual conversations without manually switching languages. Early customer feedback has been very positive — testers noted that this eliminates the frustration of manually switching languages during live events. This feature is currently in preview and will roll out to all users in January.

Improved ASR Model

Our latest multilingual ASR model — trained over 18 days — achieves our best Word Error Rate (WER) to date, covering Japanese, English, Chinese, Korean, and Vietnamese. Additional improvements include:
  • Increased custom phrase dictionary capacity from 150 to 500 entries with space delimiter support
  • Support for constrained language candidates (e.g., limit recognition to Japanese and Vietnamese only)
  • Improved audio buffer architecture with sliding-window context, reducing errors at chunk boundaries

Neural Speaker Diarization

Deployed a new neural speaker diarization model to production, replacing the previous clustering-based approach. The new model provides more accurate speaker separation, especially in longer meetings. We also began training Japanese-specific speaker diarization models to further improve accuracy for Japanese meetings.

Speaker Identification Research

Began evaluating advanced speaker identification models to improve automatic speaker labeling in transcripts. This research will enable more accurate “who said what” attribution in multi-speaker meetings.

New Features

HENNGE One SSO Integration

Enterprise single sign-on via HENNGE One is now available across web, iOS, and Android. With SSO enabled, users are automatically redirected to their organization’s HENNGE authentication when signing in. This feature is ideal for organizations that require centralized identity management.

Subtitle Translation Dictionary

The caption/subtitle feature now supports custom translation dictionaries. You can define specific translation mappings that are applied during real-time subtitle translation — ensuring industry-specific terminology, product names, and proper nouns are translated correctly every time.

Data Intelligence Improvements

  • New sales analysis template that automatically extracts client information from meeting transcripts (company size, industry, language, location, interest level)
  • Custom prompt support for mind map node creation
  • Improved CJK (Chinese/Japanese/Korean) text processing

Signup Experience Update

The registration form now includes optional fields for your intended use case and region, helping us tailor the onboarding experience. We’ve also added a country/region field for better localization support.

Mobile App Updates

  • iOS and Android updates published with all December improvements
  • Five UI fixes: Meeting title application, placeholder text, default message length, file upload button label, and AI suggestion button positioning
  • SSO redirect fix: Mobile deep links now properly redirect through SAML SSO authentication
  • New settings screen: Added a settings gear icon on the home screen, with custom phrase and dictionary management moved into a dedicated settings area

Bug Fixes

  • Fixed translation stuck on short messages: Very short utterances (e.g., “Yeah they do.”) could fail to trigger translation. The system now handles single-event messages correctly
  • Fixed plan downgrade notification duplication: Downgrading then cancelling could produce duplicate notifications
  • Fixed language detection regression: Language detection was no longer correctly swapping the translation pair when it detected the target language being spoken
  • Fixed signup data not saving: Registration questionnaire responses were briefly not being stored — resolved with a same-day hotfix
  • Fixed bilingual mode message duplication: Resolved a timing issue where text correction and language switching could produce duplicate messages
  • Fixed React state API call loop: Custom phrase API was hitting rate limits due to a state management issue — resolved immediately
  • Fixed listener mode corruption: Resolved an issue where listener mode state could become corrupted during passive meeting participation
  • Fixed dictionary modal rendering: The dictionary management modal was appearing behind other UI elements

UI/UX Improvements

Pricing Page Refresh

  • Corrected all plan descriptions with accurate feature specifications
  • Updated terminology: “guest” changed to “listener” across all languages
  • Standardized AI accuracy display across all plan tiers
  • Added missing per-month indicators on minute displays
  • Added Business Intelligence feature pointer to each plan card

Other UI Updates

  • Invoice settings: Billing tab renamed to “Invoice setting” — email, phone number, and name can now be updated directly from web and mobile
  • Plan usage display: Now shows plan name with STT limit, bold text for total usage, and a billing confirmation button
  • Subtitle timestamp drift fixed for more accurate playback synchronization
  • Presentation mode now automatically hides dictionary and notes panels for a cleaner display
  • Removed “beta” labels from all features
If you have any questions, please contact us via this form.
## Official Links