Container Images (release 240725)
deepgram/onprem-api:release-240725
Equivalent image tag to deepgram/onprem-api:1.118.7
deepgram/onprem-engine:release-240725
Equivalent image tag to deepgram/onprem-engine:3.73.4
Minimum required NVIDIA driver version: >=525.60.13
deepgram/onprem-license-proxy:release-240725
Equivalent image tag to deepgram/onprem-license-proxy:1.6.2
deepgram/onprem-billing:release-240725
Equivalent image tag to deepgram/onprem-billing:1.9.3
deepgram/onprem-dgtools:release-240725
Equivalent image tag to deepgram/onprem-dgtools:2.1.7
This Release Contains The Following Changes
Enables transcription of mixed English and Spanish audio, using our new English/Spanish multilingual code-switching model.
Requires a new language model; contact your Deepgram representative.
Invoke with the URL parameter language=multi.
(Beta) Enables entity detection of 50+ supported entity types for English pre-recorded audio. Previously, this feature has only been available on our hosted endpoint.
Requires a new entity-detection model; contact your Deepgram representative.
In your api.toml file, enable with the feature flag entity_detection = true.
Invoke with the URL parameter detect_entities=true.
(Beta) Expands our redaction capabilities to support 50+ supported entity types for English pre-recorded audio.
Requires a new entity-detection model; contact your Deepgram representative.
In your api.toml file, enable with the feature flags entity_detection = true and entity_redaction = true.
Invoke with the URL parameter redact={class_to_redact}.
(Beta) Enables next-generation streaming entity formatting for English audio. Formats entities for maximum readability and naturalness, such as dates, times, emails, and mailing addresses.
Requires a new entity-detection model; contact your Deepgram representative.
Requires a non-formatted English streaming speech-to-text model. Appropriate models contain the substring en.non-formatted.streaming in their name.
In your engine.toml file, enable with the feature flag streaming_ner_formatting = true.
Invoke with the URL parameter smart_format=true.
Logs a warning if a user-provided TOML configuration file path is invalid and the default fallback configuration is used.
Reduces Engine container logging noise by downgrading log-level of the /models endpoint heartbeat from INFO to DEBUG.
Fixes an error when summarizing empty audio.
Returns empty audio instead of a 400 error when unutterable text is provided for TTS.
Keeps our software up-to-date.