Shunya LabsShunya LabsPlayground

Horizon · Japanese-English

Japanese-English Speech to Text

Convert conversations that naturally move between Japanese and English into structuredtext for meetings, interviews, customer conversations, podcasts, media, voiceapplications, search, analytics, and AI workflows. Use Shunya Horizon Universalfor Japanese and English speech, and evaluate Horizon Code-Switch forconversations where both languages naturally appear inside the same interaction.

Japanese + English inside one real conversationHorizon Universal for Japanese and English speechHorizon Code-Switch where both languages naturally appearStructured transcript Search / LLM / CRM / Analytics
Transcript · Japanese-English

この meeting は tomorrow に変更できますか?

Japanese + English inside one real conversation
MeetingsInterviewsCustomer conversationsVoice applications

Japanese-English speech to text

Japanese + English inside one real conversation

Business and technology conversations do not always stay in one language. Shunya’s broader Horizon platform supportsboth Japanese and English speech recognition, while Horizon Code-Switch is designed for conversationsthat naturally change languages within the same sentence or interaction.

A speaker might say

この meeting は tomorrow に変更できますか?

or

Payment status を確認してください。

or

新しい API integration について説明します。

Basic flow

Japanese + English audioHorizonStructured transcriptSearch / LLM / CRM / Analytics / Voice Application / Automation

Japanese and English can appear in

Inside one real conversation

  • Meetings
  • Customer support
  • Technology discussions
  • Product conversations
  • Interviews
  • Podcasts
  • Global teams
  • Voice applications

Why mixed-language recognition matters

Language switching can happeninside one sentence

An English-only recognition path needs to handle the Japanese speech surrounding those words. HorizonCode-Switch is instead designed for speech that changes languages inside the same interaction.

Consider

次の meeting は Friday に変更します。

A Japanese-only recognition path needs to handle

meeting and Friday

  • meeting
  • Friday

Important support note

Japanese-English pair supportshould be verified

However, it does not explicitly name Japanese + English as a dedicated production Code-Switch pair.Confirm supported pairs in the deployment you plan to use.

Horizon supports

  • Japanese Speech to Text
  • English Speech to Text
  • Horizon Universal
  • Multilingual ASR
  • Horizon Code-Switch
  • Mixed-language transcription
  • Multilingual meetings
  • Multilingual podcasts
  • Multilingual video

Safe positioning

Japanese-first speechHorizon UniversalEnglish-first speechHorizon UniversalJapanese-English mixed conversationVerify current Code-Switch pair supportTest production audioDeploy appropriate configuration

Try it yourself

Test the Japanese-English your usersactually speak

Try examples such as:

Try

Meeting は tomorrow の 10 AM です。

Try

Payment confirmation がまだ届いていません。

Try

この account の status を確認してください。

Try

API integration は next week に開始します。

Try

Order number 48217 を check してください。

Test with

your users actually speak

  • Global meetings
  • Interviews
  • Support calls
  • Technical discussions
  • Product demonstrations
  • Podcasts
  • Voice notes
  • Phone recordings
  • Names
  • Acronyms
  • Product terminology

Why mixed-language recognition matters

Traditional multilingual processing may try

Horizon Code-Switch is instead designed for speech that changes languages inside the same interaction.

Traditional multilingual processing may try

Detect, change model, then merge

  • Detect Japanese
  • Process Japanese
  • Detect English term
  • Change model
  • Merge output

Horizon Code-Switch

Speech that changes languages inside the same interaction

  • Language A
  • Language B
  • Horizon Code-Switch
  • Continuous transcript

English technical vocabulary

Japanese business speech can containEnglish-derived terminology

Production evaluation should include the actual vocabulary used by the business.

Example

API integration の status を確認します。

Common categories include

English-derived terminology

  • API
  • Login
  • Payment
  • Account
  • Meeting
  • Status
  • Order
  • Support
  • Cloud
  • Server
  • Report

Japanese writing system

Mixed Japanese text can contain several scripts

Written Japanese commonly combines Kanji, Hiragana, and Katakana inside the samelanguage. Japanese-English conversations may therefore combine Kanji, Hiragana, Katakana, and Latin-script English terminology.

Written Japanese commonly combines

Kanji, Hiragana, and Katakana

  • Kanji
  • Hiragana
  • Katakana

Japanese-English conversations may therefore combine

Japanese scripts and Latin-script English

  • Kanji / Hiragana / Katakana
  • Latin-script English terminology

Japanese writing system

Script handling is an important partof transcription quality

This makes script handling an important part of mixed-language transcription quality.

Example

Japanese speechJapanese textEnglish product or technical vocabulary

Horizon Code-Switch

Built for conversations thatchange languages

Horizon Code-Switch is designed for conversations that naturally switch betweenlanguages inside the same sentence or interaction.Verify the current configuration before making a pair-specific deployment.

Important

Code-Switch handles mixed speech

Code-Switch handles mixed speech.

Important

Japanese + English is not named as a production pair

It does not specifically name Japanese + English as a supported production pair.

Mixed speech

Language ALanguage BHorizon Code-SwitchContinuous transcript

Real-time Japanese-English transcription

Process speech while the useris still speaking

Horizon supports real-time speech recognition through WebSocket streaming.For mixed Japanese-English speech, validate the correct model routing beforedeployment.

Live audio

Microphone / application streamHorizonInterim transcriptFinal transcriptApplication

Use for

while the user is still speaking

  • Voice applications
  • Agent assist
  • Live meetings
  • Live captions
  • Interactive AI
  • Customer support
  • Voice search

Streaming audio

Support live and telephony-oriented streams

Horizon supports live audio representations including int16 PCM, float32, μ-law, and A-law.This supports both general real-time audio and telephony-oriented workflows.

Live audio representations

int16 PCM, float32, μ-law, A-law

  • int16 PCM
  • float32
  • μ-law
  • A-law

Live audio

Live audioWebSocketHorizonTranscript

Batch transcription

Process completed recordings

For recorded speech, Horizon returns a complete transcript, then optional intelligence.

Audio

Meeting / podcast / interview / callHorizonComplete transcriptOptional intelligence

Best for

For recorded speech

  • Recorded meetings
  • Interviews
  • Podcasts
  • Webinars
  • Customer calls
  • Research
  • Media
  • Post-call analysis

Streaming vs batch

Live and completed audio are separate workflows

Exact Japanese-English mixed-language routing should be verified against the active Code-Switch configuration.

RequirementStreamingBatch
Live Japanese-English
Recorded Japanese-English
Partial Transcript
Complete Transcript
Voice Applications
Live Meetings
Post-call Intelligence
WebSocket
HTTP API

Customer support

International support conversationsmay mix Japanese and English

The STT layer provides speech understanding. The business system remains responsible for actual account and customer actions.

Common terms may include

Account, Login, Payment, Order, Refund, Ticket, Support, Password, Status

  • Account
  • Login
  • Payment
  • Order
  • Refund
  • Ticket
  • Support
  • Password
  • Status

Customer audio

Customer audioHorizonStructured transcriptCRM / Search / QA / Analytics

Voice applications

STT becomes the listening layer

Horizon provides machine-readable input for search, LLMs, agents, CRM, analytics,and automation. For dedicated Japanese-English Code-Switch behavior, validate deployed pair support.

User speaks

Japanese / English / mixed speech

Horizon

Horizon

Transcript

Transcript

LLM / Agent

LLM / Agent

Application response

Application response

Machine-readable input for

Search, LLMs, agents, CRM, analytics, and automation

  • Search
  • LLMs
  • Agents
  • CRM
  • Analytics
  • Automation

Where the mix shows up

Japanese and English inside global conversations

Global meetings

Japanese-English switching is common in international teams

The meeting-transcription source positions Horizon Universal for global meetings and Horizon Code-Switch for mixed-language meetings.

  • 今日の agenda を確認します。
  • The deployment starts next Monday.
  • 次に API migration について話します。

Meeting → Appropriate Horizon model → Transcript → Speakers → Timestamps → Search / Notes / AI

Interviews

Convert bilingual interviews into searchable text

Possible use cases

  • Journalism
  • Customer research
  • Hiring
  • Product interviews
  • Academic research
  • Media production

Interview → Horizon → Transcript → Speaker labels → Timestamps → Search / Analysis

Podcasts

Japanese belongs to the global podcast workflow

Horizon Universal includes Japanese for global-language podcasts. For mixed-language podcasts, it separately recommends Horizon Code-Switch. Japanese is explicitly listed under global podcast coverage. Mixed-language routing should follow the current supported configuration.

  • Japanese / English / mixed speech

Japanese-English podcast → Horizon Universal / Code-Switch as appropriate → Transcript → Speakers → Timestamps → Summary → Publishing / Search / AI

Speaker diarization

Preserve who said what

Horizon supports speaker diarization across its broader platform.

SPEAKER_00

次の meeting は Friday です。

SPEAKER_01

Okay, I will update the schedule.

SPEAKER_00

ありがとうございます。

Use for

Meetings, interviews, podcasts, customer calls, panels, and research

  • Meetings
  • Interviews
  • Podcasts
  • Customer calls
  • Panels
  • Research

Japanese-English meeting workflow

A practical international-teamarchitecture

This follows the architecture supported in the meeting transcription workflow.

Japanese-English meetingHorizonSpeaker-aware transcriptWord timestampsSummary / Intent / KeytermsLLMAction items / Decisions / Follow-upsBusiness systems

Speaker identification

Resolve known speakerswhere configured

Registered voice profiles can be combined with diarization to identify known speakers. Unknown speakers canremain anonymous labels. Speaker identification should not be treated as legal identity verification.

Before

SPEAKER_01

I'll update the schedule.

After

Agent Priya

I'll update the schedule.

Word-level timestamps

Connect transcript words to audio

Verbose transcription output can include start time, end time, and confidence for individual words.

FieldVerbose output
Start time
End time
Confidence

Search

APITranscript matchTimestampAudio

Use for

Meeting playback, search, podcast editing, captions, video editing, call review, and QA

  • Meeting playback
  • Search
  • Podcast editing
  • Captions
  • Video editing
  • Call review
  • QA

Confidence scores

Flag difficult transcript regions

Mixed Japanese-English conversations may contain names, acronyms, product names, technical terminology, addresses,and numbers. Confidence scores can support review. They should not be treated as guarantees of correctness.

Mixed Japanese-English conversations may contain

Confidence scores can support review

  • Names
  • Acronyms
  • Product names
  • Technical terminology
  • Addresses
  • Numbers

They should not be treated as guarantees of correctness.

Numbers and references

Structured details in the conversation

Production evaluation should include the actual vocabulary used by the business.

Example

Order number 48217 を check してください。

Audio input

Use common recorded formats

Horizon batch transcription supports WAV, MP3, M4A, OGG, FLAC, and WebM.

Batch

Documented audio formats

  • WAV
  • MP3
  • M4A
  • OGG
  • FLAC
  • WebM
Audio fileHorizonStructured transcript

This supports

Meetings, interviews, podcasts, voice notes, and customer recordings

  • Meetings
  • Interviews
  • Podcasts
  • Voice notes
  • Customer recordings

Video transcription

The video format does not choosethe speech model

The language and conversation determine the model - not the MP4, MOV, or other container. For multilingual video,use the appropriate multilingual/Code-Switch workflow. Horizon supports multilingual and mixed-language video transcription. Visual information requires separate visual analysis when needed.

Shunya's video workflow supports

Submit supported video files to batch transcription

  • MP4
  • MKV
  • MOV
  • AVI
  • WebM

Japanese-English video workflow

VideoAudio extractedAppropriate Horizon modelTranscriptTimingSpeakersSearch / Editing / Captions / AI

Multilingual meeting intelligence

Transcription can become structured meeting data

The meeting workflow supports optional speaker identification, emotion, intent,sentiment, summary, keyterms, and redaction.

Speaker identification

Emotion

Intent

Sentiment

Summary

Keyterms

Redaction

Meeting

MeetingTranscriptSpeakersTimingIntelligenceLLM / Business Systems

Possible downstream extraction

Decisions, action items, owners, deadlines, and follow-up questions

  • Decisions
  • Action items
  • Owners
  • Deadlines
  • Follow-up questions

Intent detection

Intent detection

Understand what a speaker wants. Intent is part of Shunya's optional intelligence layer.

  • Account issue
  • Order status
  • Product question
  • Meeting request
  • Support request
  • Human-agent request

Transcript → Intent → Workflow

Sentiment analysis

Sentiment analysis

Add post-call context. Shunya supports sentiment analysis among optional transcription intelligence features. Sentiment should supplement the transcript rather than replace it.

  • Customer support
  • QA
  • Research
  • Sales review
  • Customer experience

Emotion analysis

Emotion analysis

Add another conversation signal. Emotion detection is part of the broader Horizon intelligence stack. Emotion should be treated as inferred metadata.

  • Support analysis
  • Coaching
  • Research
  • Media analysis

Summarisation

Summarisation

Reduce long bilingual conversations. Summarisation is part of Shunya's supported optional intelligence layer.

  • Main topic
  • Questions
  • Decisions
  • Product terminology
  • Customer concern
  • Resolution
  • Next steps

Audio → Transcript → Summary → Search / CRM / Knowledge Base

Keyterms

Keyterms

Preserve important domain terminology. Keyterm handling is available in Shunya's batch intelligence workflows.

  • Product names
  • APIs
  • Brand names
  • Technical vocabulary
  • Company names
  • Domain terms

Raw transcript → Keyterm processing → Consistent downstream terminology

Redaction

Redaction

Mask configured sensitive terms. Shunya supports custom redaction within supported transcription workflows.

  • Support
  • CRM
  • QA
  • Analytics
  • Internal archives

Sensitive information → Configured redaction → Masked transcript

Developers

Add Japanese and English speech recognition to an application

The broader Horizon API supports recorded audio through HTTP transcription and liveaudio through WebSocket transcription. POST /v1/audio/transcriptions is supportedacross the meeting and video workflows. Use the current supported Code-Switchrouting only after confirming pair availability.

transcriptions.txt

RECORDED AUDIO
HTTP transcription

BATCH ENDPOINT
POST /v1/audio/transcriptions

Configure

Model, Language, Output format, Diarization, Speaker identification, Word timestamps

  • Model
  • Language
  • Output format
  • Diarization
  • Speaker identification
  • Word timestamps
  • Intent
  • Sentiment
  • Emotion
  • Summary
  • Redaction
Explore the Speech to Text API

Choose the right model

The model follows the speech

Shunya supports this model-family approach across Voice to Text workflows.

Japanese-first

Horizon Universal

Use the appropriate Horizon Universal configuration for Japanese.

English-first

Horizon Universal

Use Horizon Universal.

Mixed Japanese + English

Horizon Code-Switch

Horizon Code-Switch where the active configuration supports the pair. Use the current supported mixed-language/Code-Switch route only after verifying pair support.

Clinical speech

Appropriate specialised model

Japanese and English on one STT platform

One speech architecture across global languages

This means a product can support Japanese and English without maintaining entirely separate speech platforms.Shunya positions Horizon Universal for broad multilingual transcription across global languages.For mixed-language speech, the Code-Switch family should be evaluated where supported.

Horizon includes

Inside the broader Horizon platform

  • Japanese Speech to Text
  • English Speech to Text
  • Korean Speech to Text
  • Chinese Speech to Text
  • Spanish Speech to Text
  • French Speech to Text
  • Other global languages

Global speech

Global speechHorizon UniversalStructured transcript

Best for

The global-language recognition path

  • Japanese speech
  • English speech
  • International meetings
  • Global customer conversations
  • Global media
  • International teams

Japanese-English is not translation

Mixed-language transcriptionpreserves what was said

A transcription workflow should first preserve that utterance as speech-to-text output. It should not automaticallyturn the whole sentence into only Japanese or only English unless a translation step is intentionally added.

Input

Payment status を確認してください。

Separate functions.

STT

Speech → text

Translation

Text → another language

  • Only Japanese
  • Only English

Evaluate before production

Test Japanese-English audio from your workflow

Customer calls, meetings, phone audio, product vocabulary, and background noise shoulddrive your evaluation.

Test real audio

Evaluate using production speech

  • Japanese-English customer calls
  • Meetings
  • Phone audio
  • Product terminology
  • Names
  • Numbers
  • Multiple speakers
  • Background noise

Pair-support caution

General Code-Switch support does notprove every pair

Horizon Code-Switch handles speech that changes language within the same conversation.Confirm that your deployed configuration supports Japanese-English before you go live.

Safe production process

Collect Japanese-English audioCheck current pair supportSelect supported model/configurationEvaluate transcript qualityDeploy

Built for production

A production Japanese-Englishworkflow may need

A production Japanese-English workflow may need

Japanese STT, English STT, and Horizon Universal

  • Japanese STT
  • English STT
  • Horizon Universal
  • Mixed-language Code-Switch evaluation
  • Streaming transcription
  • Batch transcription
  • Speaker diarization
  • Speaker identification
  • Word-level timestamps
  • Confidence scores
  • Meetings
  • Podcasts
  • Interviews
  • Video
  • Intent
  • Sentiment
  • Emotion
  • Summarisation
  • Keyterms
  • Redaction
  • API integration
  • Self-hosted deployment

Requirement | Horizon

RequirementSupport
Japanese Speech Recognition
English Speech Recognition
Horizon Universal
Code-Switch Platform
Dedicated Japanese-English PairVerify active configuration
Streaming Transcription
Batch Transcription
Speaker Diarization
Speaker Identification
Word-level Timestamps
Confidence Scores
Multilingual Meetings
Global Podcasts
Multilingual Video
Intent Detection
Sentiment Analysis
Emotion Analysis
Summarisation
Keyterms
Redaction
API Integration
Self-hosted Deployment
Japanese + English speechHorizonSupported multilingual / Code-Switch routeStructured transcriptSearchCRMAnalyticsLLMVoice ApplicationKnowledge Base

Questions

Frequently Asked Questions

What is Japanese-English Speech to Text?

It converts conversations containing Japanese and English speech into structured written text.

Does Shunya support Japanese Speech to Text?

Yes. Japanese is explicitly included in the broader Horizon multilingual/global language network.

Does Shunya support English Speech to Text?

Yes. English is supported by Horizon Universal.

Does Shunya support Code-Switch?

Yes. Horizon Code-Switch is designed for speech that naturally changes languages inside the same sentence or interaction.

Is Japanese + English listed as a dedicated Code-Switch pair?

Confirm Japanese + English as a supported production pair in your deployment.

Should I verify pair support before deployment?

Yes. The current Code-Switch configuration should be checked before making a dedicated Japanese-English routing.

Which model should I use for Japanese-only speech?

Use the appropriate Horizon Universal configuration for Japanese.

Which model should I use for English-only speech?

Use Horizon Universal.

Which model should I use for mixed Japanese-English speech?

Use the current supported mixed-language/Code-Switch route only after verifying pair support.

Can Japanese-English speech be transcribed in real time?

Horizon supports real-time WebSocket transcription. Exact mixed-pair routing should be validated.

Can recorded Japanese-English audio be transcribed?

Horizon supports batch transcription for recorded audio. Mixed-language configuration should be confirmed.

Can Japanese-English meetings be transcribed?

The platform supports global and multilingual meetings, while Code-Switch is used for mixed-language meetings.

Can Japanese-English podcasts be transcribed?

Japanese is explicitly part of Horizon Universal's global podcast workflow, while mixed-language podcasts can use Code-Switch where supported.

Can video be transcribed?

Yes. Shunya supports video transcription by extracting the audio track from supported containers.

Can multiple speakers be separated?

Yes. Speaker diarization can label transcript segments by speaker.

Can known speakers be identified?

Yes. Registered voice profiles can be combined with speaker identification.

Can word-level timestamps be returned?

Yes. Verbose output can include start time, end time, and confidence.

Can intent be detected?

Yes. Intent is part of the optional intelligence layer.

Can sentiment be analysed?

Yes. Sentiment analysis is supported.

Can emotion be analysed?

Yes. Emotion detection is supported.

Can conversations be summarised?

Yes. Summarisation is supported in applicable batch workflows.

Can keyterms be processed?

Yes. Keyterm handling is available in batch transcription workflows.

Can sensitive terms be redacted?

Yes. Custom redaction is supported.

Does Shunya publish a Japanese-English-specific WER?

Shunya does not publish a dedicated Japanese-English word error rate. Test it on your own audio.

Does the 3.10% OpenASR figure apply to Japanese-English?

No. It should not be re-labelled as Japanese-English Code-Switch performance.

Is Japanese-English STT the same as translation?

No. STT converts speech into text. Translation changes the language and is a separate workflow.

Get started

Turn Japanese + English conversationsinto structured text

Global teams do not always stay in one language. A meeting may begin in Japanese. A product term may be English. Atechnical discussion may move between both. Use Horizon Universal for Japanese and English speech, and evaluatethe Code-Switch route for mixed Japanese-English conversations. Test real meetings, interviews, podcasts, customerconversations, product terminology, names, acronyms, multiple speakers, and real production audio.