Skip to content
Model infoModelsSnapshot: 15 Oct 2026

IBM / Watson next-generation STT

Watson STT: Hindi (Indian) Telephony

IBM speech recognition for telephone audio.

Transcription
availableClosed weightsSpeech-to-text
Official documentation (opens in a new tab)Reviewed 14 Oct 2026
14 of 24 specification fields collected
File upload limit100 MB synchronous/streaming request; 1 GB asynchronous jobSee specification notes below
Transcription outputTextSee specification notes below
Input billingSee ratesIBM Watson Speech to Text
Output billingSee ratesPublished units and conditions below

Specifications

Provider-documented facts. “Not collected” means we have not verified that field yet, not that the feature is unsupported.

Context / input limit
Not collected
Maximum output
Not collected
Input modalities
AudioSource for Input modalities (opens in a new tab)
Output modalities
TextSource for Output modalities (opens in a new tab)
Reasoning
Not collected
Tool calling
Not collected
Structured output
Not collected
Streaming
SupportedSource for Streaming (opens in a new tab)

Live audio and transcription over WebSocket. HTTP synchronous output arrives after processing the request.

Batch processing
SupportedSource for Batch processing (opens in a new tab)

Asynchronous recognition jobs, not a token Batch API or a discounted batch tariff.

Fine-tuning
Not collected

Language-model customization is recorded separately. It does not establish a general weight fine-tuning API.

Knowledge cutoff
Not collected
Release date
Not collected
Parameters
Not collected
Architecture
Deep neural recognition with bidirectional audio analysisSource for Architecture (opens in a new tab)

The documentation links RNN-transducer research. Exact deployed parameter count and checkpoint architecture were not verified.

Languages
Hindi (Indian)Source for Languages (opens in a new tab)
Context caching
Not collected
Maximum audio file
100 MB synchronous/streaming request; 1 GB asynchronous jobSource for Maximum audio file (opens in a new tab)

At least 100 bytes. Duration depends on codec and compression; these are byte limits, not a fixed context duration.

Audio formats
WAV, FLAC, MP3 / MPEG, Ogg Opus / Vorbis, WebM Opus / Vorbis, L16 PCM, A-law, µ-law / basic, G.729Source for Audio formats (opens in a new tab)
Live stream limits
WebSocket frame limit: 4 MB; multiple recognition requests can share a connectionSource for Live stream limits (opens in a new tab)
Concurrent requests
Plus plan: 100 simultaneous transcription requests across all interfacesSource for Concurrent requests (opens in a new tab)
Included free allowance
Lite plan: 500 recognition minutes/month, shared across models; no customizationSource for Included free allowance (opens in a new tab)

Service-instance allowance, not a separate grant per model. Inactive Lite instances may be deleted after 30 days.

Usage metering
Audio duration billing; silence included; monthly aggregate roundingSource for Usage metering (opens in a new tab)
Audio sample rates
Minimum 8 kHz audioSource for Audio sample rates (opens in a new tab)

Higher rates are resampled. This is not a parameter count or a model context window.

Customization options
Language customization: GA; grammars: GA; no acoustic-model customizationSource for Customization options (opens in a new tab)

Paid plan required. Product-version migration dates remain in the linked provider table.

Access & pricing

Prices belong to an endpoint and billing tier. A provider listing does not guarantee access for every account or region.

IBM Watson Speech to Text

Provider source (opens in a new tab)
API model ID
hi-IN_Telephony
Access
Pass this exact identifier in the model parameter. REST synchronous, REST asynchronous and WebSocket access share the model selection.
Regions
Use the service-instance URL for the provisioned region. Product-edition and voice-specific availability apply; no universal residency guarantee inferred.
routes
POST /v1/recognize, POST /v1/recognitions (asynchronous), WebSocket /v1/recognizeSource for routes (opens in a new tab)

Transport behavior is documented separately by the HTTP and WebSocket sources.

Request limits
100 MB sync/streaming request; 1 GB async; minimum 100 bytesSource for Request limits (opens in a new tab)
jobRetention
Asynchronous results expire after one week by default; results_ttl can overrideSource for jobRetention (opens in a new tab)
lowLatency
SupportedSource for lowLatency (opens in a new tab)
acousticCustomization
Not supportedSource for acousticCustomization (opens in a new tab)

Prices retain their published currencies and billing units below.

Published billing units · original currency · observed 14 Oct 2026
CategoryUsageRateUnit
Speech recognitionPlus: below 1M minutes/month1$0.02audio minute
Speech recognitionPlus: 1M+ minutes/month2$0.01audio minute

Rate conditions

  1. Plus plan: aggregate monthly usage of 1–999,999 audio minutes. Includes silence/noise. Monthly aggregate rounded to nearest minute, not each API call. No discount for asynchronous requests is stated.
  2. Plus plan: aggregate monthly usage of at least 1,000,000 audio minutes. This rate applies to all minutes for that month, not just incremental minutes above the threshold. Includes silence/noise. Monthly aggregate rounded to nearest minute, not each API call. No discount for asynchronous requests is stated.
Pricing source (opens in a new tab)

Data & enterprise terms

Retention, training use, regional processing and enterprise terms have not been summarized for this endpoint.

Caching

Caching can reuse earlier input processing. Availability, lifetime and billing depend on the endpoint. A cache hit still generates a new response.

01 / INITIAL INPUTProcess & retain

The system processes eligible input and retains reusable state.

02 / LATER REQUESTCheck available state

Reuse depends on matching input, session state and the cache policy.

03 / CACHE HITReuse & generate

Reusing state can reduce repeated work. Any charges follow the billing rules below.

IBM Watson Speech to Text

hi-IN_Telephony
Not verified

No provider-priced inference cache contract verified. Client-side audio reuse is not provider prompt caching.

How it works
Not collected

No provider-priced inference cache contract verified. Client-side audio reuse is not provider prompt caching.

Minimum reusable input
Not collected

No provider-priced inference cache contract verified. Client-side audio reuse is not provider prompt caching.

Lifetime & refresh
Not collected

No provider-priced inference cache contract verified. Client-side audio reuse is not provider prompt caching.

How to enable it
Not collected

No provider-priced inference cache contract verified. Client-side audio reuse is not provider prompt caching.

Usage fields
Not collected

No provider-priced inference cache contract verified. Client-side audio reuse is not provider prompt caching.

Billing rules
Not collected

No provider-priced inference cache contract verified. Client-side audio reuse is not provider prompt caching.

Limits & data boundaries
Not collected

No provider-priced inference cache contract verified. Client-side audio reuse is not provider prompt caching.

No verified cache rate for this endpoint. An empty rate is not a zero charge.

What the measurements say

No benchmark observations have been added for this model. We do not turn a vendor description into a performance score.

Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.

The record behind the model

Why it is included

IBM documents this proprietary Watson speech service model or managed voice family. The service exposes recognition/voice selectors, not downloadable trained checkpoints. Closed-weight classification is inferred for the product model; public Granite Speech research releases and referenced open research implementations are separate.

Classification evidence (opens in a new tab)

Sources reviewed