01
Speak
Say the message, note, question, instruction, or conversation naturally.
Horizon · Voice to Text
Turn spoken words into accurate, structured text for voice notes, dictation, customer conversations, meetings, voice agents, search, and real-time AI applications. Speak naturally and let Shunya Horizon convert your voice into text - with real-time transcription, speaker labels, word timestamps, multilingual recognition, and conversation intelligence.
Hi, I wanted to check whether my payment was processed yesterday and when I should expect the confirmation.
Real-time transcription · HorizonVoice to Text
Speaking is often faster and more natural than typing. But useful voice-to-text needs to do more than capture individual words. Real speech includes pauses, accents, corrections, interruptions, background noise, multiple speakers, and sometimes more than one language. Horizon is built to convert that real-world voice into structured text.
204+
Languages supported across the Horizon platform
3.10%
Composite WER across 8 OpenASR benchmarks
Sub-500ms
Streaming transcription for real-time voice applications
#1
OpenASR composite leaderboard performance
| Capability | Horizon · Voice to Text |
|---|---|
| Live Speech | |
| Recorded Voice | |
| Streaming Transcription | |
| Batch Transcription | |
| Automatic Language Detection | |
| Speaker Diarization | |
| Speaker Identification | |
| Word-level Timestamps | |
| Confidence Scores | |
| Multilingual Recognition | |
| Code-switching | |
| Conversation Intelligence |
Try it yourself
Don't test voice recognition with a carefully written sentence. Try a real thought, customerrequest, voice note, product name, address, number, or multilingual sentence.
Example:
Hi, I wanted to check whether my payment was processed yesterday and when I should expect the confirmation.
Speak, don't type
Typing requires you to stop and translate a thought into a keyboard interaction. Voice lets the thought happen first.Voice to Text turns that spoken information into written text that can then be searched,edited, analysed, stored, or sent into another application.
01
Say the message, note, question, instruction, or conversation naturally.
02
Horizon converts the speech into text.
03
Add language, speakers, timestamps, and optional intelligence.
04
Send the text into notes, search, analytics, an LLM, a CRM, a database, or another workflow.
Voice → text
Real-time voice transcription
Some voice-to-text workflows happen after a recording is complete. Others need the transcript immediately. Horizon supports real-timestreaming transcription for applications that need to listen and respond while speech is happening.
Sub-500ms
Streaming transcription latency
Partial transcripts
Receive interim text while the speaker is still talking.
Final transcripts
Finalise transcript segments as the speech progresses.
Real-time applications
Streaming flow
The application does not need to wait for the entire conversation to end before using the text.
Real voices aren't perfect
Voice recognition has to work with the way people actually communicate.
People stop to think in the middle of sentences.
A speaker may begin a sentence, change direction, and start again.
The same language can sound very different across regions and speakers.
Speech may come from offices, traffic, homes, public spaces, or phone lines.
Spoken conversations contain abbreviations, shortened phrases, slang, and conversational grammar.
A sentence may begin in one language and finish in another.
Calls and meetings can contain interruptions and rapid speaker changes.
Voice to Text needs to recognise conversation - not just carefully dictated speech.
Who said what
When more than one person is speaking, a plain block of text loses part of the conversation.Speaker diarization separates the transcript according to speaker.
Example
SPEAKER_00 Hi, I'm calling because I haven't received my order yet.
SPEAKER_01 I can check that for you. Could you share your order number?
SPEAKER_00 Sure, it's 48217.
Instead of returning the entire conversation as one block, the transcript preserves the conversational structure.
Useful for
Speaker identification
When registered voice profiles are configured, anonymous speaker labels can also be resolved against known speakers.
Voice conversation → structured transcript
Know when it was said
Voice becomes much easier to work with when the transcript knows where every word came from.Horizon can return timing and confidence information for individual words.
Multilingual voice
Real conversations can change languages without warning. This is common in India, multilingualcustomer support, international teams, and everyday conversations.
Hinglish
मेरा payment yesterday process हुआ था but confirmation अभी तक नहीं आया.
Tanglish
Order confirm ஆயிடுச்சு but delivery address கொஞ்சம் change பண்ணணும்.
Benglish
Paymentটা complete হয়েছে but confirmation message এখনও পাইনি.
A traditional speech workflow may need separate language detection and model switching.Horizon includes specialised code-switching models designed for mixed-language conversations.
One speaker · mixed languages
Native mixed-language recognition without forcing the speaker to choose one language for the entire conversation.
From voice to intelligence
A conversation may contain useful information beyond the transcript itself. For supported batch workflows,Horizon's intelligence layer can add structured information to the same transcription request.
Who spoke when.
Resolve registered voices into speaker identities.
What does the speaker want?
Is the conversation positive, neutral, or negative?
How does the speaker sound during different segments?
What happened in the conversation?
Standardise important business and domain terminology.
Return translated transcript output where required.
Mask profane language.
Hide defined sensitive terms and phrases.
Know exactly where individual words occurred.
Structured conversation data
The spoken conversation becomes structured information that other systems can use.
Voice notes
Voice notes are convenient because they are quick to create. They become less convenient when you have to listento every recording again to find one sentence. Voice to Text can convert spoken notes into written information.
Speak an idea
Follow up with the design team tomorrow about the new onboarding flow and confirm whether the mobile screens are ready.
Get the text
Follow up with the design team tomorrow about the new onboarding flow and confirm whether the mobile screens are ready.
Then use it
Voice in. Text out. The information becomes easier to reuse once it is no longer trapped inside an audio recording.
Real-world use cases
Convert a customer's speech into text in real time so an AI agent can understand the request.
Turn spoken thoughts, notes, reports, and instructions into written text.
Transcribe calls for agent assistance, records, search, analytics, summaries, and quality workflows.
Capture conversations as searchable text with speaker and timing information.
Convert spoken interviews into structured transcripts for research, hiring, journalism, and analysis.
Let users speak a query instead of typing it.
Provide written output for spoken information and voice-first experiences.
Turn lectures, discussions, and spoken explanations into searchable written material.
Capture information by voice when typing on a keyboard or touchscreen is inconvenient.
Convert human voice into machine-readable text before passing the request to an LLM or agent.
Voice agents
A voice agent needs to understand the user before it can generate a response. Voice to Text is the first intelligence layer in that interaction.
User speaks
Can you check the status of my order?
Horizon
“Can you check the status of my order?”
User hears
Your order has been shipped and should arrive tomorrow.
Speech → text → intelligence → speech
The conversation remains voice-first while the application processes the interaction as structured text.
Recorded + live voice
Voice-to-text applications do not all operate in real time. Horizon supports both workflows.
Streaming
Use streaming transcription when the text is needed while the speaker is still talking.
Best for:
Batch
Process completed recordings when immediate transcription is not required.
Best for:
| Requirement | Streaming | Batch |
|---|---|---|
| Live voice | ||
| Recorded voice | ||
| Partial transcripts | ||
| Complete transcripts | ||
| Voice agents | ||
| Conversation intelligence | ||
| WebSocket | ||
| HTTP API |
Developers
Horizon provides one speech-recognition platform for live and recorded voice workflows. Capture microphone or application audio through a WebSocket connection for partial and final transcription, or send recordings through the transcription API with optional intelligence.
{
"language_code": "auto",
"file": "meeting.wav",
"enable_diarization": true
}Live voice
Capture microphone or application audio, send it through a WebSocket connection, and receive partial and final transcription.
Recorded voice
Send a recording through the transcription API, configure language and optional intelligence, and receive structured transcript output.
Configure
Output
Basic workflow
Choose the right voice model
One speech model does not need to handle every use case in exactly the same way. The Horizonfamily includes specialised models for different speech environments.
Model
Broad multilingual transcription across global languages, with particularly strong English performance.
Model
Purpose-built recognition for Indian languages, regional speech, and Indic conversational audio.
Model
Designed for conversations that naturally switch between languages inside the same sentence or interaction.
Model
Speech recognition designed for healthcare terminology and clinical conversations.
Choose for the conversation
Built for production
Adding a microphone icon to an application is easy. Building reliable voice recognition behind that button is harder. Productionapplications need real-time transcription, multilingual recognition, speaker awareness, structured output,privacy controls, APIs, and the ability to operate on the audio users actually produce.
| Requirement | Horizon |
|---|---|
| Voice to Text | |
| Real-time Voice | |
| Recorded Voice | |
| Multilingual Recognition | |
| Automatic Language Detection | |
| Code-switching | |
| Speaker Diarization | |
| Speaker Identification | |
| Word-level Timestamps | |
| Intent Detection | |
| Sentiment Analysis | |
| Emotion Detection | |
| Summarisation | |
| Redaction | |
| Streaming API | |
| Batch API | |
| Self-hosted Deployment |
Voice → application
Voice becomes an input method that the rest of the software stack can understand.
Questions
Voice to Text is speech-recognition technology that converts spoken words into written text.
Yes. Horizon supports both recorded speech transcription and real-time streaming transcription.
Yes. Horizon supports real-time ASR through a WebSocket streaming workflow.
Yes. Streaming speech recognition can be used to build dictation, voice-note, captioning, and other voice-input experiences.
Yes. Speaker diarization can separate transcript segments according to speaker.
Registered voice profiles can be combined with speaker diarization to resolve recognised speakers where speaker identification is configured.
Yes. Verbose transcription output can contain word-level start time, end time, and confidence information.
Yes. Horizon supports multilingual speech recognition, and the specialised Horizon Code-Switch model is designed for mixed-language conversations.
Yes. Horizon Code-Switch is designed for conversations where Hindi and English naturally appear together.
The Horizon platform supports automatic language detection when a language is not explicitly locked.
Optional batch intelligence capabilities include intent detection, sentiment analysis, emotion detection, summarisation, and other transcript enrichment features.
Yes. Custom keyword redaction and profanity masking can be enabled in supported transcription workflows.
Yes. Horizon provides HTTP APIs for recorded transcription and WebSocket APIs for streaming speech.
Shunya supports self-hosted deployment for organisations that require greater control over speech data and infrastructure.
Get started
Speak naturally. Let Shunya handle the transcription. Try a voice note, customer request, question, name, number,mixed-language sentence, or real conversation. See how your voice becomes searchable, structured text.