Skip to main content
Detailed documentation about Speech to Text is available here, and supported languages here.
API/transcribe

Update to word-level transcript messages on /transcribe

When wordLevel is enabled in the configuration of a /transcribe session, final transcripts now deliver all finalized word items together in a single message, with data as a list matching the transcript message envelope already used by /streams. Previously, each finalized word arrived as its own singular message.Interim previews, returned when interimResults is also enabled, remain segment level and now arrive as a one-element list carrying the in-progress segment.Sessions with wordLevel disabled (default) are unaffected and keep the unchanged singular data object per segment. If you consume word-level messages, update your client to handle data as a list. See word-level transcripts in the API reference and the new word-level transcripts guide for details.
Language Model

Spanish speech to text now rated as premier

An updated Spanish (es) model is now available across all speech to text endpoints. This language is now rated as Premier with performance improvement expected across all medical domains.Evaluations across five datasets show an average relative improvement (percent decrease) in word error rate (WER) of 28.4%.This update also makes spoken punctuation and formatting available for Spanish. Further details are available on the languages and transcribe pages.
Language Model

Updated German (de, de-CH) and French (fr, fr-CH) models now available

Raw model performance evaluations (i.e., before formatting and punctuation evaluation) were conducted using 10 datasets for German and 6 for French and followed the methods outlined in our May 2026 benchmark paper. This includes use of our public MedDictate and MedTerm datasets.Results include the following average relative improvement (percent change) in word error rate (WER) and medical term recall (MTR):
  • German:
    • WER: -18.75%
    • MTR: 4.35%
  • French:
    • WER: -10.46%
    • MTR: 5.23%
    Percent change = (new value - old value)/old value * 100Improvements are indicated by a reduction in WER (lower is better) and increase in MTR (higher is better)
API/transcribe

Command updates during active connection now supported

Real-time speech to text now supports ability to update commands mid-session. Send a commands_update message to add, update, or remove commands. The server processes requests in order they are received and communicates to the client when command updates are available for recognition.Only commands can be updated mid-session. Other configuration parameters such as formatting, replacements, primaryLanguage, and keyterms cannot be changed without closing and re-establishing the WebSocket connection. See the full API specification for details.

Maintenance

  • Increase maximum session duration for /transcribe and /streams websocket connections from 2-hours to 4-hours.
  • Increased validation requirements for replacements as 10-word, 100-character maximum for find and replace values.
API

Maintenance

  • Separate handling of “hyphen” and “dash” in spoken punctuation:
    • Hyphens do not add punctuation between words ("hello-world")
    • Em-dash adds spaces between words ("hello — world")
  • Expand French and German decimal number formatting so that both “point” and “comma” are recognized.
  • Add support for German abbreviations “HWK”, “BWK”, “LWK”, and “SWK” in vertebrae formatting.
Language Model

Updated Swiss German (dialectical) model now available

An updated dialectical Swiss German (gsw-CH) model is now available across all speech to text endpoints. This language is now rated as Premier with performance improvement expected across all medical domains. Further details are available on the languages page.
Language Model

Maintenance

  • Expanded base tier languages available for audio file processing via /transcripts endpoint. All languages are now available consistently across all speech to text endpoints. See details on the languages page.
  • Improve handling of disfluency and initialisms (fr, fr-CH)
API

Maintenance

  • Update to formatting to improve capitalization, casing, punctuation, and contextual awareness.
API

Maintenance

  • Update GET/transcripts response so that transcripts are ordered based on created_at timestamp (descending). Query parameter direction can also be added to define sort order as descending (desc) or ascending (asc)
API

Speech to text formatting service update Beta

Speech to text endpoints utilize a formatting service to properly output dates, numbers, units, measurements, and more. This service is undergoing an upgrade to broaden coverage for such use cases and improve contextual disambiguation in support of continued reliability of speech to text performance.A one month beta period will allow clients to test the new formatting model in comparison to current generally available functionality. No client action will be required to utilize the updated formatting service upon general availability.Use the following language codes to conduct beta testing:
Clients should utilize the beta period to test behavior within their application and notify Corti - contact us directly or submit feedback - with concerns or issues. An update will be noted here and in the upcoming change log when a GA date is confirmed.
API

Word level transcript responses

New configuration parameter wordLevel (object) now available for all speech to text endpoints to enable word-level transcript response messages. When enabled, transcript responses will be individual words instead of multi-word phrase transcript segments.
  • Enable this for word level timestamps in transcript output.
  • Only final transcripts return word level output; interim results are unaffected.
  • Additional wordLevel configuration options to be added in subsequent releases.
  • Full details available in api-reference for transcribe, streams, and transcripts.

Maintenance

  • Updated multichannel audio validation to support a maximum of 8 audio channels. See audio formats for full details.
API

Maintenance

  • Formatting configuration now available on /streams and /transcripts endpoints.
  • Apply fix for rawTranscriptText in transcript response messages.
  • Docs update: Clarified validation requirements for keyterms and replacements as 5-word maximum in addition to 50 character limit.
Language Model

Maintenance

  • Expanded support for formatting of {number} x {number} patterns (e.g. “five times two hundred milligrams” -> “5 x 200 mg”).
  • Improve performance of French (fr) spoken punctuation detection.
  • Swiss French (fr-CH) medications output in title case.
  • Expanded support for number recognition in Swiss French (fr-CH).
  • Add support for alternative phrasings for Dutch (nl) spoken punctuation.
  • Apply fix for diarization to improve consistency of speaker separation.
  • Apply fix for “noon” and “midnight” so they’re returned as text instead of formatted as times.
Resources

Tympany WER analysis tool now available as open source

Building with speech-to-text does not stop at transcription. You need to know where the model succeeds, where it fails, and whether each difference matters for the workflow you are building.Tympany is an open source, local-first web app for speech-to-text evaluation. Use it to create “Beyond WER” (BeWER) reports from reference and generated transcripts, review errors, recompute error rates after excluding differences that are not true model failures, and identify errors that need to be reported to Corti for ongoing STT improvement.See full details and gain access to the tool here.
Language ModelAPI/transcribe/streams

Language expansion and updates

  • Expansion of Base tier to add support for 27 new languages
  • Languages upgraded from Base to Enhanced tier - Spanish, Arabic
  • Expansion of Enhanced tier to add support for new language - Finnish
  • Expansion of Premier tier to add support for new language - English (AU)
See full details on the languages page, including ability to search by language (name or code) and filter by performance tier.Corti values the opportunity to expand to new markets, but we need your collaboration and partnership in speech-to-text validation to expand language support from base to enhanced or premier tiers. Contact us to learn more.
API/streams/transcripts

Replacements and Keyterms now available on Streams and Transcripts endpoints

The features replacements and keyterms are now supported on all three STT endpoints - transcribe, streams, and transcripts.Additionally, some configuration parameters have been updated on the streams and create transcripts endpoints for improved consistency. No timeline is currently planned to remove support for the deprecated parameters. See details in upcoming changelog.

Maintenance

  • Ensure replacements correctly handle casing and punctuation
  • Fix capitalization of first word in transcript (capitalize) and after colon and semicolon (lowercase)
  • Update formatting option locale:short to use four-digit year (e.g., 2026)
  • Increase maximum audio file duration accepted for upload recordings from 60 to 120 minutes
API/transcribe

Keyterm bias

New feature for transcribe web socket endpoint now available - define keyterms the system should be aware of and biased towards to ensure proper nouns unknown by the STT model are recognized and improve recognition reliability for other general terms. See feature details here.
Language ModelAPI/transcribe/streams/transcripts

Swiss French fr-CH now supported

Language code fr-CH is now supported for use of Swiss French speech to text.Beyond current French speech to text handling, this model has improved support for Swiss medications and returns punctuation without non-breaking spaces (as used in French (fr or fr-FR) documentation).The languages page has been updated, and detailed information for punctuation handling on text insertion is available here.
API/transcribe

Commands with wildcard variable

New dictation feature, wildcard variable commands now available. Unlike enum command variables, wildcard variables provide the ability to recognize a command based on undefined, open-ended text. A literal trigger word, such as “select”, is required before defining a wildcard variable.See feature details with examples here and configuration in the API reference.
API/transcripts

Updated asynchronous method for audio file processing

The /transcripts endpoint provides batch audio file processing using a synchronous-to-asynchronous method, where requests process synchronously for 25 seconds before hitting timeout and continuing asynchronously. A new request parameter, async, is now available to force asynchronous processing. This method returns the 202 status response right away, instead of waiting the full timeout window, so that clients can poll status endpoint and get full transcript when it is completed.See more details about /transcripts here and the updated parameter in the api-reference.
API/transcribe

Replacement Rules

New dictation feature, replacement rules, now available. Define terms to “find” and “replace” in STT output using the replacements configuration option in /transcribe requests. See details and examples here and API configuration here. Replacements configuration is limited to 1,000 items per connection.

Bug fixes

  • “D3” now correctly handled
  • Updated measurement formatting so that “0°” is supported
API/transcribe/streams

Raw PCM audio now supported

Audio processing has been updated so that raw PCM audio can be sent over web socket secure (wss) audio streaming with /transcribe and /streams. When using raw PCM audio, audioFormat must be declared in configuration. See details here.
Language Model/transcripts

Infrastructure update for /transcripts endpoint to improve speech-to-text performance for audio file processing

Infrastructure updates to improve performance of audio file processing:
  • Updates to batch sync-to-async audio file processing to ensure consistent latency (RTFx 100)
  • Model updates to reduce risk of hallucinations while improving medical term recall (English, German, French, Danish)
  • Ensure efficient GPU utilization, refine load balancing and partition management (enhance autoscaling of ASR services)
No client action is necessary to select a specific model version or use the new architecture. See full details on creating transcripts from audio files here.
API/languages

New /languages endpoint now available

New REST endpoint available for clients to programmatically retrieve speech to text languages available per endpoint. See GET/languages/.
Language Model/transcribe/streams/transcripts

Updated German and French STT models released

Update STT models for German (de, de-CH) and French (fr) have been released for improved medical term recall and WER performance. This release focused on general terminology recognition, medication names, and punctuation handling.
API/transcribe/streams

Audio Health Events

A new feature for real time audio streaming, audio health events are notifications from speech to text system when audio quality may be compromised or problematic. The feature can be enabled with boolean parameter in transcribe and streams configuration.The following two scenarios are supported in this initial release:
  • Speech Quality Issue Detected - triggered by background sounds, white noise, or compromised audio.
  • Long Silence Detected - defined as 10s of continuous audio without any sound/noise.
Additional functionality is expected in subsequent releases, as well as ability to define threshold for alerting in the API configuration.Learn more about audioEvents here.
API/transcribe/streams

Command and formatting updates

  • Added timeout to return formatted text on silence
  • Added non-breaking space before French punctuation characters
  • Added support for square meter (m²) and square centimeter (cm²) units
  • Fixed handling of decimal number format so that negative numbers are supported
  • Security hardening: Improved validation of user-supplied command phrase inputs to prevent malformed values from affecting processing
Language Model/transcribe/streams

Updated French speech to text models now available

French speech to text model (language code fr) has been updated for improved medical term recall and conversational, far field audio support.
API/streams

Transcript output formatting now supported on Streams endpoint

Formatting is now supported on the streams endpoint, with server-defined default values applied automatically. At this time, API configuration of formatting options is not supported for streams as it is for transcribe endpoint.
API/transcribe/streams
Response messages for configuration errors have been clarified. See details here and here.
Language Model/streams

Improved diarization now available

Updates to Corti’s Speech to Text models and diarizer, in which reliability and precision for speaker separation and identification have been optimized, have been released.See details on this feature here and configuration recommendations for streams and transcripts requests here.
API/transcribe

Formatting updates and improvements

  • Add formatting for more units: “g/dL”, “cL”, “hL”, “pL”, “mg/L”, “mg/kg”, “U/mL”, “mL/min”,
  • Add support for “times two” -> “x2” and “two plus” -> “2+” patterns (1 through 10)
  • Fix handling of German spaced number variations
Language Model/transcribe/streams

Next-generation speech recognition models and infrastructure

New and improved model architecture and infrastructure bring the following actions and benefits:STT models for the following languages have been updated on both /transcribe and /streams endpoints:Beyond defining the desired language code in API requests, no client action is necessary to select a specific version of a model. Please contact us for support or further information.
API/transcribe

Interim Results now available during real-time dictation for low latency transcript previews

Use the interimResults configuration parameter in /transcribe real-time dictation requests to have transcript previews returned from the server at a faster rate than final transcripts.See full feature details here, parameter configuration details here, and availability by language here.

Locale-based formatting for dates, times, and numbers

Dictation formatting can now apply proper styling based on local standards for dates, times, and numbers. Locale is determined based on the primaryLanguage defined in the web socket configuration.See full formatting details per language here and parameter configuration details here.
Note that legacy date parameter values are still supported by the API; no breaking changes were implemented.

Dictation formatting improvements

  • Expand units handling
  • Expand
  • Improve vertebrae formatting: Support for both letter number and letter number letter number patterns (e.g., “L three” -> “L3” and “L three L four” -> “L3-L4”)
  • Improve TNM and cancer stage formatting: Support for T number N number M number and stage number patterns (e.g., “T two N one M one a” -> “T2 N1 M1a” and “stage two b” -> “Stage IIB”)
  • Improve percentage handling: Move handling from spokenPunctuation to formatting so that percent symbol is only returned when there is a number measurement (e.g., dictation, “What percent sure are you question mark eighty percent” -> “What percent sure are you? 80%”)
  • Expand handling of German dates: Years without spaces (e.g. “zweitausendsechsundswanzig” -> “2026”) and new pattern (e.g., include pattern “siebter fünfter neunzehn vierundachtzig” -> “07/05/1984”)
  • Expand handling of German units: Units with -n ending (e.g., “Millilitern”)
  • Add new default for ordinals, numerals_above_nine: Ordinals one through nine are written out (first, second, third) and ten and above are abbreviated (10th, 11th, 12th)
API/transcribe

Dictation formatting improvements

  • Improved support for en-GB regional spelling variations
  • Improved support for de-CH, gsw-CH regional spelling variations
  • Remove extra whitespace observed with degree and percent symbols
  • Updated handling of single digit numbers, now represented as numeral when followed by year/ month/ week/ hour/ day/ minute/ second(s)
  • Updated handling of hyphenated numbers and age dictation
  • Updated handling of “one twenty” (and similar number pattern) dictations
  • Updated handling of Month-Year dictation
  • Fix for an issue that occasionally caused both formatted and unformatted versions of transcript text to be emitted
API/transcribe
Update to handling of number formatting in Danish.
API/transcribe
Bug fixes:
  • Update to French formatting for improved number handling
  • Update to Danish formatting for liter abbreviation handling
  • Bug fix for extra whitespace being returned with commands or punctuation
  • Command service performance and reliability improvements
API/transcribe

Dictation Formatting now supported in more languages

Dictation formatting is now available in English, German, and French. There is also limited support in Danish with additional improvements in progress.See detailed API specification here and documentation of all available formatting options here.
Language Model/transcribe/streams

Improved handling of initialisms in German dictation

An updated version of the German (de) language model was released with improved handling of dictated initialisms: spoken letters, numbers, and abbreviations. Over 800 additional medical abbreviations were introduced to the system, with focused improvement on dental tooth exams, adding support for ICD-10 codes, and ability to have individual letters/numbers returned from the model.
API/transcribe/streams

Audio input validation

Improved server-side validation of streamed audio - if audio does not meet requirements outlined here, then a 400 Invalid Audio error will be returned.
API/transcribe/streams

Addition of new web socket event: flush

Provide clients the ability to force clear the audio buffer without closing the connection. In response to a "type": "flush" message from the client, the server will return recognized text and/or commands and respond with "type": "flushed", and keep the web socket connection open.See more details on /transcribe here and /streams here.
Language Model/transcribe/streams/transcripts

Update to Swiss German language codes

There are now two different language codes that may be used for Swiss German use cases:
  • Swiss German (language code gsw-CH), where dialectical Swiss German is spoken (recommended for AI scribe use cases)
  • Swiss High German (language code de-CH), where Swiss High German is spoken (recommended for dictation use cases)
See more details here, or please contact us if you need further assistant in selecting the best language code.
Language Model/transcribe/streams

Updated Danish language model

Updated Danish (da) language model is now available for dictation (/transcribe) and ambient (/streams) use cases. The language is rated as premier tier as over 158,000 medical terms were included in the training and validation data sets.
API/transcribe
Bug fixes:
  • Fixed an issue that prevented automaticPunctuation parameter from working as expected. Furthermore, spokenPunctuation and automaticPunctuation are mutually exclusive: Only one of these parameters should be set to true in a given /transcribe configuration, and if both settings are present and set to true, then spokenPunctuation will take precedence. See more detail here.
Language Model/transcribe/streams

Updated French language model

Updated French (fr) language model is now available for dictation (/transcribe) and ambient (/streams) use cases. The language is rated as premier since over 170,000 medical terms were included in the training and validation data sets.
Language Model/transcribe

Dictation now supported in Dutch and Norwegian

Norwegian (no) and Dutch (nl) are now available for use on the /transcribe endpoint for dictation use cases. As a result they have been moved to the enhanced language tier. Additionally, Hungarian (hu) medical terminology has been expanded and improved.See more information about supported functionality for dictation here and language tiers here.
Language Model/transcripts
Feature limitation:
  • Some of the recent /transcripts endpoint language model updates (de, en, en-GB, fr) have been rolled back so that performance of asynchronous audio file processing can be improved. Infrastructure updates are underway to improve model performance.
  • Speech-to-text accuracy from /transcripts asynchronous audio file processing may be degraded as compared to real-time audio processing via the /streams and /transcribe APIs, which are not impacted by this issue.
Language Model/transcribe/streams/transcripts

Updated English language model

Updated English (en) language model is now available for dictation (/transcribe), ambient (/streams), and transcription (/transcripts) use cases. This version is rated as premier tier as over 170,000 medical terms were included in the training and validation data sets.
API/transcribe/streams/transcripts

Improved transcoding support in speech-to-text APIs

Update to both streaming and asynchronous audio file processing so that transcoding is supported. Previously audio files were required to conform to 16-bit, 16kHz formatting. Now, assuming proper file types are used, any precision and sample rate are accepted. See more details here.
API/transcripts
Parameter deprecation:
  • Definition of the parameter modelName is no longer required in /transcripts API requests. The latest and greatest model available per language will be applied automatically.
  • If the argument is included in the request, then it will be ignored. If the configuration is otherwise valid, then the request will process as expected. See full specification here.
API/transcribe
Feature limitation:
  • The interim (preview) results feature of the /transcribe API is not performing as expected. As a result, it is being disabled for most languages to prevent issues with speech-to-text latency and accuracy.
  • If the interimResults parameter is included in the request for a language that does not support this functionality it will be ignored and, so long as the request is otherwise valid, the configuration will be accepted.
Language Model/streams

Conversational transcripts now supported in Arabic

Arabic (language code ar) is now available for ambient documentation workflow - Capture conversations spoken in Arabic via the /streams API.
API/transcribe

New dictation functionality available: Formatting

New dictation functionality available: Formatting. Take control over how dates, time, units, and numbers should be transcribed in the streaming speech to text output.See detailed API specification here and documentation of all available formatting options here.
Language Model/transcribe/streams/transcripts

Updated German language model

Updated German (de) language model is now available for dictation (/transcribe), ambient (/streams), and transcription (/transcripts) use cases. This version is rated premier tier as over 150,000 medical terms were included in the training and validation data sets.
Language Model/transcribe

Dictation now supported in Hungarian

Hungarian (hu) Enhanced language model is now available for dictation (/transcribe).
Language Model/transcribe
Swedish (sv) Enhanced language model is now available for dictation (/transcribe).
Language Model/streams/transcripts
Updated Norwegian (no) and Swedish (sv) Base language models are now available for ambient documentation (/streams) and transcription (/transcripts) workflows.
Language Model/transcribe
Danish (da), German (de), and French (fr) Enhanced language models now available for dictation (/transcribe) workflows.
New API endpoint, /transcribe now available! Use this endpoint for stateless, real-time streaming dictation workflows. See more details here in the API reference and here for access to the Corti Dictation Web Component.
Language Model/streams
Swiss German (de-CH) Enhanced language model now available for ambient documentation (/streams) workflows.
Language Model/streams
French (fr) Enhanced language model now available for ambient documentation (/streams) workflows.
APILanguage Model/transcribe/streams/transcripts
Introducing a new tier system for defining functionality and performance of speech to text language models. Read more about it here.
Language Model/streams/transcripts
Updated German (de) and Swiss German (de-CH) language models available.
Language Model/streams/transcripts
Updated Danish (da) and Swedish (sv) language models available.
APILanguage Model/streams/transcripts
Announcing the launch of Corti AI Platform. Read more about it here.The following languages are supported by Corti speech to text: English (en for US English, and en-GB for UK English), Danish (da), German (de), Swiss German (de-CH), French (fr), Swedish (sv), Spanish (es), Norwegian (nl), Dutch (no), Italian (it), and Portuguese (pt)