July 2020 Virtual appliances 3.5.0, Batch container 7.0.0, Real-time container 1.1.0

Appliance Download

If you have any questions about our latest release, our Support team will be happy to help.

Tag

UPDATE

Deployment

BATCH CONTAINER
REAL-TIME CONTAINER
BATCH VIRTUAL APPLIANCE
REAL-TIME VIRTUAL APPLIANCE
CLOUD OFFERING

Feature

Global English accuracy increase

This release sees the latest iteration of our Global English language pack which means our best-in-class transcription capability just got better. We have applied new language training techniques which have shown up to 10% relative gains in Word Error Rate accuracy over the previous version in our testing.

UPDATE

BATCH CONTAINER
REAL-TIME CONTAINER
BATCH VIRTUAL APPLIANCE
REAL-TIME VIRTUAL APPLIANCE
CLOUD OFFERING

Global Spanish

In this release, our Spanish has gone global featuring more accents and dialects than ever before. We have applied new language training techniques that – in our tests – have shown up to 10% relative gains in WER accuracy over the previous version.

You shouldn’t have to adjust the way you speak so that speech recognition systems can understand you, let us worry about that. With Global Spanish, you don’t have to worry about deploying multiple accent-specific Spanish language packs – multiplying operational costs, complexity and hosting space – to get the best transcription output. Global Spanish is a single Spanish language pack trained on data drawn from global sources especially those from the LATAM region, to deliver the first global Spanish model in the market.

Speechmatics’ Global Spanish language model is now even more robust when it comes to the challenge around background noise such as music or chatter. Our internal testing has shown a noticeable gain in transcription accuracy providing the voice in the foreground has good clarity.

Note: There is no change to the language code to use the Global Spanish language model. Users should continue to use the ‘ES’

NEW

BATCH CONTAINER
REAL-TIME CONTAINER
REAL-TIME VIRTUAL APPLIANCE

Session start-up optimization when using Custom Dictionary  

Custom Dictionary enables users to add their own vocabulary to any languages supported by Speechmatics. It means that users can add words or phrases that are unique to their use case to increase the probability of that word being transcribed correctly.

From this release we now enable users to cache their Custom Dictionary word lists. Why? Often customers who use the Custom Dictionary feature will repeatedly use the same word list for their transcription jobs and real-time sessions. With users often taking advantage of the full Custom Dictionary 1,000-word limit, this can sometimes slow down the transcription starting especially for real-time.

In our tests, the ability to cache the Custom Dictionary has shown an approximate 90% reduction in start-up time on our latest real-time appliance for a word list of 1,000 words.

  • Batch and Real-time ContainersSpeechmatics provides a caching mechanism that ensures that the container is able to write a Custom Dictionary object to a customer-owned and operated shared space so that it can be reused by other sessions in future.Caching behaviour is also configurable dependant on size. For example, behaviors can be set to only cache the Custom Dictionary that is of a certain size (bytes) to manage the capacity of the cache. Note: Users are responsible for the management for their own Custom Dictionary caches. Note: Users are responsible for the management for their own Custom Dictionary caches.
  • Real-time Virtual Appliance – The virtual appliance will contain the space for the Custom Dictionary word lists to be cached. Speechmatics manage the size of custom dictionaries that can be cached in this area.
    • You can cache multiple dictionaries in the same language. The only limit is the total disk size of the appliance (between 100-300MB depending on appliance size).
    • The size of the Custom Dictionary has also shrunk, meaning that our smallest virtual appliance can store just over 1,200 different custom dictionaries with 1,000 words each and our largest (Plus) virtual appliance, almost 4,000.
    • If the cache space is full the Custom Dictionary that hasn’t been submitted for the longest time will be deleted to make space.

NEW

BATCH CONTAINER
BATCH VIRTUAL APPLIANCE

SRT transcript output support

SRT is now available as a supported output format for transcription files being processed using the Speechmatics Batch Virtual Appliance and Container (joining the capability of the Cloud offering that started support for SRT earlier in July). This simplifies the steps for creating subtitles and captions from any Speechmatics batch product.

The Speechmatics SRT uses industry-standard and BBC guidelines for the default parameters which limit the amount of characters/words per subtitle. These standards are up to 37 characters per line and 2 lines per chunk of text.

Users can modify the defaults within the configuration object up to 100 characters per line and 10 lines per chunk of text.

Check out our SRT demo for more information.

FYI

BATCH VIRTUAL APPLIANCE

Upcoming deprecation of the V1 API within the next Batch Virtual Appliance release

The legacy V1 API that the Batch Virtual Appliance currently supports will be discontinued in our next release in October 2020. We will align the product with the same V2 API used by the Speechmatics Cloud offering.  To familiarize yourself with the V2 API we recommend that you take a look at the cloud offering documentation. However, if you need additional information please reach out to the support team.

V1 API will be a supported mechanism in the 3.5 release however the following 3.6 release (currently scheduled for October 2020) will only support the new V2 API. 3.5 will still be supported via N-1. This means that users do not have to upgrade if they are not to do so. However, by the 3.7 release 3.5  falls out of support along with the ability to use the V1 API.

Release timeline

N V1 API supported V2 API supported N-1 V1 API supported V2 API supported
Jul 2020

3.5

Yes

No

3.4

Yes

No

Oct 2020

3.6

No

Yes

3.5

Yes

No

Jan 2021

3.7 No Yes 3.6 No

Yes

July 2020 Cloud Offering

Documentation

If you have any questions about our latest release, our Support team will be happy to help.

Tag

NEW

Deployment

CLOUD OFFERING

Feature

SRT (SubRip) subtitle format support in the cloud offering

SRT is now available as a supported egress format for transcription files being processed using the Speechmatics cloud offering. This simplifies the steps for creating subtitles or captions within the Speechmatics cloud offering.

The Speechmatics SRT use industry-standard and BBC guidelines for the default parameters which limit the amount of characters/words per subtitle. These standards are up to 37 characters per line and 2 lines per chunk of text.

Users can modify the defaults within the configuration object up to 100 characters per line and 10 lines per chunk of text.

Check out our SRT demo for more information.

April 2020 Virtual Appliance 3.4, Batch Container 6.3, Real-time Container 1.0, Cloud offering

Appliance download

If you have any questions about our latest release, our Support team will be happy to help.

Tag

NEW

Deployment

CLOUD OFFERING

Feature

New Speechmatics cloud service instance hosted in the US

The Speechmatics cloud service is now hosted in the United States in addition to Europe. This delivers better data control for US cloud customers by ensuring their data remains within their geographical region. More information on the cloud offering hosted in the US is available in the news section.

NEW

BATCH APPLIANCE
REAL-TIME APPLIANCE

New appliance option for real-time and batch virtual appliances

The new ‘plus’ virtual appliance option delivers full language parity across all deployment types (cloud offering, containers and virtual appliances) and enable customers to leverage all of Speechmatics’ languages no matter their deployment type.

The ‘plus’ virtual appliances include the following languages: Global English (en), Mandarin (cmn), Norwegian (no), Arabic (ar), Bulgarian (bg), Czech (cs), Greek (el), Finnish (fi), Hungarian (hu), Croatian (hr), Lithuanian (lt), Latvian (lv), Romanian (ro), Slovak (sk), Slovenian (sl), Turkish (tr), Malay (ms).

NEW

REAL-TIME CONTAINER

Real-time Container

Real-time is now available within a container! The demand for real-time transcription has increased and so we want to ensure customers have the flexibility to deploy this capability. Customers can now deploy a standalone container for any Speechmatics language that can perform low-latency, real-time transcription using the same V2 WebSocket Speech API that the Real-time Virtual Appliance provides. More information on the Real-time Container can be found in the news section, additional technical details are available in the real-time datasheet.

NEW

BATCH CONTAINER
REAL-TIME CONTAINER
BATCH VIRTUAL APPLIANCE
REAL-TIME VIRTUAL APPLIANCE

Consistent transcription of numbers less than 10 as words in Global English (en)

Historically, our number recognition has been good, but we know that the format in the transcript is not always consistent. Sometimes there might be a digit while other times a word. In the latest release, there have been improvements to ensure that all numbers less than ten are consistently transcribed as words. There is no need to make any changes as this is a default within the product. Additional information on this feature can be found in the news section.

UPDATE

BATCH CONTAINER
REAL-TIME CONTAINER
BATCH VIRTUAL APPLIANCE
REAL-TIME VIRTUAL APPLIANCE

Speechmatics has increased the number of terms that can be mapped between American and British English when using the ‘Output Locale’ feature within Global English

Global English is a unique single English language pack which supports the ‘Output Locale’ feature. This can be used by customers to enable the transcription output to be formatted to represent either British or American spellings of words. This increases geographic accuracy and reduces the effort of human users of the transcripts.

NEW

BATCH VIRTUAL APPLIANCE
REAL-TIME VIRTUAL APPLIANCE

HTTPS support for virtual appliances

HTTPS can now be enabled. Customers can configure their own certificates on both Batch and Real-time Virtual Appliances so that all API traffic (speech, management and monitoring) is secure.

Customers can be confident that they are communicating with a trusted appliance and that all information sent to and received from the appliance is encrypted.

April 2020 Cloud Offering

Documentation

If you have any questions about our latest release, our Support team will be happy to help.

Tag

NEW

Deployment

CLOUD OFFERING

Feature

Consistent transcription of numbers less than 10 as words in English

Historically, we have been pretty good at number recognition but we know that the format in the transcript is not always consistent. Sometimes there might be a digit while other times a word. In the latest release, there have been improvements to ensure that all numbers less than ten are consistently transcribed as words. Don’t worry there is no need to make any change as this is a default within the product.

Tag

NEW

Deployment

CLOUD OFFERING

Speechmatics has significantly increased the number of terms that can be mapped between US and British English when using the ‘Export Locale’ feature within Global English

Global English is a unique single English language pack which supports the ‘Output Locale’ feature. This can be used by the customer to enable the transcription output to be formatted to represent either British or American spellings of words increasing geographic accuracy and reducing the effort of human users of the transcripts.

March 2020 Cloud offering

Documentation

If you have any questions about our latest release, our Support team will be happy to help.

Tag

BUG FIX

Deployment

CLOUD OFFERING

Feature

Fixed: When a customer requests a file after it has been purged we will now return a 410 “File is gone”

Previously when a requesting a file that had been purged from the cloud offering customers would receive a 500 error. This wont happen anymore. Now means that customers who are requesting files which no longer exist because it has been purged or rejected they will get an accurate response of a 410 status code that the file has gone instead of an error code.

Other fixes

We made a number of additional internal changes too. You might not be able to see them but trust us they’re there meaning we can continue to deliver you best in class transcription from our cloud. For more information, you can check out our latest release notes.

February 2020 Virtual Appliance 3.3, Batch Container 6.2, Cloud offering

Appliance Download

If you have any questions about our latest release, our Support team will be happy to help.

Tag

NEW

Deployment

CLOUD OFFERING
BATCH CONTAINER

Feature

Speechmatics now supports automatic speech recognition for the Malay language

Malay extends our existing language set to 31 available languages. Like all of our languages, Malay delivers some of the lowest word error rate results validated and benchmarked against other prominent ASR providers.

Malay comes equipped with Advanced Punctuation which supports periods, commas, exclamation marks and question marks. These characters enhance readability and deliver transcription closer to natural speech.

Malay is available through the Speechmatics V2 cloud offering or as a container using the language code ‘ms’.

UPDATE

CLOUD OFFERING
BATCH APPLIANCE
REAL-TIME APPLIANCE
BATCH CONTAINERS

English & Spanish language packs improved for better accuracy in telephony use cases

We’ve added spontaneous 16kHz and 8kHz speech training data into both English and Spanish language packs. This will help to improve recognition for use cases like meetings and phone calls. Additionally, this delivers more robustness for recognition in quieter audio use cases. This is available through the Speechmatics V2 cloud offering, the Batch or Real-Time Virtual Appliance or as a container.

NEW

BATCH CONTAINERS

Option to leverage additional available CPU’s to accelerate transcription speed for containers

The new ‘parallel’ feature enables container customers to take advantage of multiple CPU resources that have been assigned to a container. This allows files over 5 minutes in duration to be split into 5-minute chunks and be processed in parallel, increasing the speed of transcription. More information is available in the Batch container v6.2.0 API and Quick Start guides.

NEW

CLOUD OFFERING
BATCH CONTAINERS

Speechmatics now supports Turkish for use in automatic speech recognition

Speechmatics can now deliver automated speech recognition of Turkish. Turkish has been benchmarked against other ASR providers to ensure it delivers leading levels of accuracy. The new language comes equipped with Advanced Punctuation. Turkish is available through the Speechmatics V2 cloud offering or as a container, using the language code ‘tr’. For more information on our Turkish language please see the Speechmatics documentation.

UPDATE

BATCH CONTAINERS

Simplified API interface to the Batch container

Currently the Batch container supports both the V1 and V2 API. To simplify the interface and to align the product with the same V2 API used by the Speechmatics cloud offering, the legacy V1 API and related output formats will be discontinued in afuture release.

We recommend that customers familiarize themselves with the configuration object used to specify job configurations which is within the Speechmatics cloud offering documentation as new features will only be supported using this mechanism.

Future notices will be provided to announce the sunset of the V1 API, along with detailed instructions on migrations to the V2 API.

UPDATE

CLOUD OFFERING
BATCH APPLIANCE
REAL-TIME APPLIANCE
BATCH CONTAINERS

Updated Dutch language

Our new Dutch language model has been trained on more data making it better than ever. The data includes the addition of Flemish and telephony training data (we’ve increased the acoustic training data by 50% and the language training data by x6).

Our Dutch model has also been extended to include enhanced punctuation functionality using our latest feature, Advanced Punctuation. The new Dutch model is available through the Speechmatics V2 cloud offering, the Batch or Real-Time Virtual Appliance or as a container using language code ‘nl’.

UPDATE

CLOUD OFFERING
BATCH APPLIANCE
REAL-TIME APPLIANCE
BATCH CONTAINERS

 

Advanced Punctuation available in Danish

Our new Danish language model benefits from Speechmatics’ industry leading punctuation. In addition to advanced punctuation characters, Danish also benefits from better capitalization. This further enhances the readability and usability of Danish in real world applications.

We have also made an upgrade to our pronunciation dictionary for numbers. The result of this is improved number recognition accuracy.

The new Danish model is available through the Speechmatics V2 cloud offering, the Batch or Real-Time Virtual Appliance or as a container using language code ‘da’.

December 2019 Cloud offering

Documentation

If you have any questions about our latest release, our Support team will be happy to help.

Tag

NEW

Deployment

CLOUD OFFERING

Feature

Speechmatics now supports Turkish for use in automatic speech recognition

Turkish extends our existing language set to 30 available languages. Like all of our languages, Turkish has been benchmarked against other prominent automatic speech recognition (ASR) providers to ensure it delivers leading levels of accuracy.

The new language comes equipped with Advanced Punctuation which supports periods, commas, exclamation marks and question marks.

Turkish is available through the Speechmatics V2 cloud offering cloud service via our API, using the language code ‘tr’. For more information on our Turkish language please see the Speechmatics documentation.

Advanced Punctuation is now available in Danish

Our new Danish language model benefits from Speechmatics’ industry leading punctuation, supporting periods, commas, exclamation marks and question marks. In addition to these punctuation characters, Danish also benefits from better capitalisation. This further enhances the readability and usability of Danish in real world applications.

We have also made an upgrade to our pronunciation dictionary for numbers. The result of this is improved number recognition accuracy.

The latest version of Danish is available through the Speechmatics V2 cloud offering. It will soon be deployable via a single container or as part of either a ‘Batch’ or Real-time’ virtual appliance using the language code ‘da’ in the next release of these deployment types. For more information on our Danish language please see the Speechmatics documentation.

August 2019 Virtual appliance 3.2, Batch container 6.1

Appliance Download

If you have any questions about our latest release, our Support team will be happy to help.

Tag

NEW

Deployment

BATCH VIRTUAL APPLIANCE
REAL-TIME VIRTUAL APPLIANCE

Feature

Advanced Punctuation

Speechmatics has released its first heavyweight punctuation model to the market. Built on over 2.5bn words, Advanced Punctuation delivers a rich set of punctuation characters as well as intelligent placement and capitalisation to improve the accuracy and readability of the transcription output.

After the success of the early access testing earlier this year, Advanced Punctuation is now available in Batch and Real-time for our English, German, Spanish and French language packs. We’ve extended the number of punctuation characters to include full stops, commas, exclamation marks and question marks, with new punctuation modelling providing our most accurate character placement to date.

Advanced Punctuation is easily configurable, now supporting an inclusion list for characters. Simply select the punctuation characters you wish to be included in your transcription output.

Note: punctuation parameters are controlled via the V2 API.

NEW

REAL-TIME VIRTUAL APPLIANCE

V2 API for Real-time

Speechmatics is changing the API that controls the Real-time appliance to align with our Batch and cloud offering. The new API enables instant access and control over new features including Advanced Punctuation and Low Latency Finals.

For more information on the Real-time V2 API, check out the API guide available in the support portal.

BETA*

REAL-TIME VIRTUAL APPLIANCE

Speaker Change (Beta)

In Speechmatics’ last release, we made Speaker Change available as a *Beta feature so that our customers could trial it and provide feedback. Therefore, some customers may have tried the feature, but we know that it needs a little more work before it’s ready for production.

If you were one of the customers that tried the Speaker Change *Beta feature in our last release, we’d love to hear how it performed. Please send your feedback to our Products team.

October 2019 Cloud offering

Documentation

If you have any questions about our latest release, our Support team will be happy to help.

Tag

NEW

Deployment

CLOUD OFFERING

Feature

Output Locale

The Speechmatics cloud offering now supports alternate spellings for transcriptions from Global English (en) using a configuration setting within the API. You can now choose British English (en-GB), US English (en-US) or Australian English (en-AU) spellings for your English transcripts.

Code example:

{

“type”: “transcription”,​

“transcription_config”: {​

“language”: “en”,​

“output_locale”: “en-GB”​

}​

}

BETA*

CLOUD OFFERING

Speaker Change (Beta)

Speaker Change is now available within the Speechmatics cloud offering. The feature uses tokens within the transcript to identify when the speaker has changed. The token is shown as ‘speaker_change’ within the JSON output and as a line break in the TXT output. You can set the sensitivity of Speaker Change as a value from 0-1. The default is 0.4, which is a good balance for most use cases. If you want to change the sensitivity, we recommend setting it between 0.3 and 0.6.

Code example:

{

“type”: “transcription”,​

“transcription_config”: {​

“diarization”: “speaker_change”,​

“speaker_change_sensitivity”: 0.55​

}​

}

UPDATE

CLOUD OFFERING

Improved Dutch including Flemish and Advanced Punctuation

Our new Dutch language model has been trained on more data making it better than ever. The data includes the addition of Flemish and telephony training data (we’ve increased the acoustic training data by 50% and the language training data by x6).

Our Dutch model has also been extended to include enhanced punctuation functionality using our latest feature, Advanced Punctuation. It is available through the API interface (using language code nl). More information is available in Speechmatics’ documentation

BUG FIX

CLOUD OFFERING

Fixed: This update resolves an issue which caused longer than expected transcription times when processing long audio files. If you only run small audio files through the system, you may not see a difference. However, tests have shown that there is a significant speed-up for media files longer than 10 minutes.

Fixed: An issue with the Custom Dictionary that affected time codes for words has also been addressed for the following languages: English (en), German (de), Spanish (es) and French (fr).

August 2019 Virtual appliance 3.1, Batch container 6.0

Appliance Download

If you have any questions about our latest release, our Support team will be happy to help.

Tag

NEW

Deployment

BATCH VIRTUAL APPLIANCE
REAL-TIME VIRTUAL APPLIANCE
BATCH CONTAINER
CLOUD OFFERING

 

Feature

Speechmatics’ latest v6 language models are now available for the following languages:

Catalan (ca), Danish (da), Dutch (nl), French (fr), German (de), Global English (en), Hindi (hi), Italian (it), Japanese (ja), Korean (ko), Polish (pl), Portuguese (pt), Russian (ru), Spanish (es), Swedish (sv).

BETA*

BATCH VIRTUAL APPLIANCE
REAL-TIME VIRTUAL APPLIANCE
BATCH CONTAINER

Speaker Change Beta*

A token added to the output transcript identifies when the speaker has changed. Using the token enhances transcript readability by clearly showing speaker boundaries.

Speaker Change is disabled by default. To enable this feature please contact our Support team.

UPDATE

REAL-TIME VIRTUAL APPLIANCE

Improved real-time latency  

We’ve simplified the configuration of latency for our real-time system, resulting in a faster, more accurate transcription. This also includes the deprecation of dynamic transcripts.

NEW

REAL-TIME VIRTUAL APPLIANCE

Real-time confidence scores  

Word-based confidence scores are now available in real-time. Confidence scores provide a probability that the transcription of each utterance matches a value within the model that each language pack has been trained on.

NEW

BATCH CONTAINERS

New languages 

We’ve extended our Global reach of languages to support Mandarin and Norwegian. If you’d like to do an evaluation, please get in touch.

*Beta features are available for customers to experiment with and consume themselves or in conjunction with customers in limited production use cases. Speechmatics does not commit to any improvement or future delivery of these features but welcomes feedback. There is no guarantee of any backward compatibility of these features. Using these features does not impact the support agreement for the software, but fixes or improvement requests specific to the Beta features are not covered within any SLAs and may or may not be considered for a future release.

     

X