Audio durationSync upload: at most 120 seconds of audio. Streaming sessions: at most 300 seconds; idle audio gap at most 60 seconds. Async file-duration ceiling not established.See specification notes below
API supports partial and committed transcripts; exact per-language checkpoint routing is not published.
Batch processing
Not collected
No reviewed model-specific contract; unknown does not mean unsupported.
Fine-tuning
Not collected
No reviewed model-specific contract; unknown does not mean unsupported.
Reasoning
Not collected
No reviewed model-specific contract; unknown does not mean unsupported.
Context caching
Not collected
No reviewed model-specific contract; unknown does not mean unsupported.
Base model
Not collected
No reviewed model-specific contract; unknown does not mean unsupported.
Published price
No reviewed public inference tariff. The account-balance API reports NGN for Nigerian accounts and USD for other countries, but a balance or sample balance is not a per-minute, character, token or question price.Source for Published price (opens in a new tab)
Homepage: 63 total languages and 12 bilingual models plus a Kinyarwanda/English/French trilingual model. Current API table: 61 unique codes, including 11 bilingual English pair routes and one trilingual route.Source for Language coverage (opens in a new tab)
Counts and grouping differ across sources; no missing checkpoint names or language routes are invented.
Streaming encoding and defaults are separately documented by the streaming source.
Audio duration limits
Sync upload: at most 120 seconds of audio. Streaming sessions: at most 300 seconds; idle audio gap at most 60 seconds. Async file-duration ceiling not established.Source for Audio duration limits (opens in a new tab)
Streaming limits apply to the connection, not every file endpoint.
Request limits
Streaming chunks: 1–32 KB. Sync: 30 requests/minute. Async: 60 requests/minute. Streaming: 3 new connections/minute and 3 concurrent connections.Source for Request limits (opens in a new tab)
The reviewed API exposes language or workflow controls, not an immutable model selector.
Access
POST https://infer.voice.intron.io/file/v1/upload/sync; Bearer API key; multipart file and use_language_asr_input. At most 120 seconds of audio; 30 requests/minute. A 503 processing timeout can return a file ID for polling.
Regions
Hosted processing region not established. Offline local deployment is a separately described commercial option, not proof of the hosted API location.
Endpoint pricing has not been collected. Check the provider for applicable rates.
Data & enterprise terms
The linked June 2024 privacy policy includes related websites, apps and APIs, including speech data. It permits service operation and improvement, and sharing for support, analytics, internal research and other stated business purposes. Retention depends on purpose and legal requirements; no universal TTL or zero-retention promise is established. Transcribe App terms preserve customer-data ownership and license processing for service operation, legal duties and written permission. A specific model-training opt-out, complete subprocessor regions and newer customer-specific terms remain unresolved.
The reviewed API exposes language or workflow controls, not an immutable model selector.
Access
POST https://infer.voice.intron.io/file/v1/upload; Bearer API key. Asynchronous file processing with status polling; 60 requests/minute. The sync 120-second audio cap is not applied to this route.
Regions
Hosted processing region not established. Offline local deployment is a separately described commercial option, not proof of the hosted API location.
Endpoint pricing has not been collected. Check the provider for applicable rates.
Data & enterprise terms
The linked June 2024 privacy policy includes related websites, apps and APIs, including speech data. It permits service operation and improvement, and sharing for support, analytics, internal research and other stated business purposes. Retention depends on purpose and legal requirements; no universal TTL or zero-retention promise is established. Transcribe App terms preserve customer-data ownership and license processing for service operation, legal duties and written permission. A specific model-training opt-out, complete subprocessor regions and newer customer-specific terms remain unresolved.
The reviewed API exposes language or workflow controls, not an immutable model selector.
Access
WSS wss://infer.voice.intron.io/stt/v1/stream; Bearer Authorization header. Select use_language_asr_input; send INPUT_AUDIO_CHUNK and COMMIT. Partial results can change before final commitment.
Regions
Hosted processing region not established. Offline local deployment is a separately described commercial option, not proof of the hosted API location.
Endpoint pricing has not been collected. Check the provider for applicable rates.
Data & enterprise terms
The linked June 2024 privacy policy includes related websites, apps and APIs, including speech data. It permits service operation and improvement, and sharing for support, analytics, internal research and other stated business purposes. Retention depends on purpose and legal requirements; no universal TTL or zero-retention promise is established. Transcribe App terms preserve customer-data ownership and license processing for service operation, legal duties and written permission. A specific model-training opt-out, complete subprocessor regions and newer customer-specific terms remain unresolved.
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.
Intron Voice API / enterprise
Not verified
No model-specific inference-cache contract verified. Stored files, streaming buffers and account balances are not prompt caching.
How it works
Not collected
No model-specific inference-cache contract verified. Stored files, streaming buffers and account balances are not prompt caching.
Minimum reusable input
Not collected
No model-specific inference-cache contract verified. Stored files, streaming buffers and account balances are not prompt caching.
Lifetime & refresh
Not collected
No model-specific inference-cache contract verified. Stored files, streaming buffers and account balances are not prompt caching.
How to enable it
Not collected
No model-specific inference-cache contract verified. Stored files, streaming buffers and account balances are not prompt caching.
Usage fields
Not collected
No model-specific inference-cache contract verified. Stored files, streaming buffers and account balances are not prompt caching.
Billing rules
Not collected
No model-specific inference-cache contract verified. Stored files, streaming buffers and account balances are not prompt caching.
Limits & data boundaries
Not collected
No model-specific inference-cache contract verified. Stored files, streaming buffers and account balances are not prompt caching.
No verified cache rate for this endpoint. An empty rate is not a zero charge.
Caching reviewed 16 Oct 2026
Intron Voice API / enterprise
Not verified
No model-specific inference-cache contract verified. Stored files, streaming buffers and account balances are not prompt caching.
How it works
Not collected
No model-specific inference-cache contract verified. Stored files, streaming buffers and account balances are not prompt caching.
Minimum reusable input
Not collected
No model-specific inference-cache contract verified. Stored files, streaming buffers and account balances are not prompt caching.
Lifetime & refresh
Not collected
No model-specific inference-cache contract verified. Stored files, streaming buffers and account balances are not prompt caching.
How to enable it
Not collected
No model-specific inference-cache contract verified. Stored files, streaming buffers and account balances are not prompt caching.
Usage fields
Not collected
No model-specific inference-cache contract verified. Stored files, streaming buffers and account balances are not prompt caching.
Billing rules
Not collected
No model-specific inference-cache contract verified. Stored files, streaming buffers and account balances are not prompt caching.
Limits & data boundaries
Not collected
No model-specific inference-cache contract verified. Stored files, streaming buffers and account balances are not prompt caching.
No verified cache rate for this endpoint. An empty rate is not a zero charge.
Caching reviewed 16 Oct 2026
Intron Voice API / enterprise
Not verified
No model-specific inference-cache contract verified. Stored files, streaming buffers and account balances are not prompt caching.
How it works
Not collected
No model-specific inference-cache contract verified. Stored files, streaming buffers and account balances are not prompt caching.
Minimum reusable input
Not collected
No model-specific inference-cache contract verified. Stored files, streaming buffers and account balances are not prompt caching.
Lifetime & refresh
Not collected
No model-specific inference-cache contract verified. Stored files, streaming buffers and account balances are not prompt caching.
How to enable it
Not collected
No model-specific inference-cache contract verified. Stored files, streaming buffers and account balances are not prompt caching.
Usage fields
Not collected
No model-specific inference-cache contract verified. Stored files, streaming buffers and account balances are not prompt caching.
Billing rules
Not collected
No model-specific inference-cache contract verified. Stored files, streaming buffers and account balances are not prompt caching.
Limits & data boundaries
Not collected
No model-specific inference-cache contract verified. Stored files, streaming buffers and account balances are not prompt caching.
No verified cache rate for this endpoint. An empty rate is not a zero charge.
Caching reviewed 16 Oct 2026
What the measurements say
Code-switched speech: Afrikaans WER
Intron · provider-reported
26.45%
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
Published homepage benchmark table data, parsed as literals without executing website code. Lower WER is better. Dataset composition, evaluator configuration and sample counts are not fully specified; not independently reproduced. Version-specific row, not a universal accuracy guarantee. Missing baseline cells stay missing; averages across unequal language sets are not compared.
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
Commercial Sahara model family described by its provider alongside proprietary training and reserved service technology. No exact Sahara checkpoint was found among the reviewed public research weights. Family names and benchmark versions do not establish immutable API selectors.
