@nhtio/adk

Appearance

3 min read · 529 words

@nhtio/adk/batteries/specialists ​

Environment-neutral aggregate barrel for bundled specialist batteries.

Remarks ​

Aggregate barrel for the specialists domain: on-device specialist models — speech-to-text (STT), OCR, and image captioning — that run alongside the LLM in the same runtime, Node and browser, rather than delegating to a cloud endpoint. There is no cloud-hosted specialist engine by design: the whole point of this domain is local media understanding a BYO executor can run without another network hop, mirroring how @nhtio/adk/batteries/embeddings favors environment-neutral, locally-runnable engines.

This barrel re-exports each modality's own aggregate barrel in full:

  • @nhtio/adk/batteries/specialists/stt — speech-to-text (transformers.js/Whisper today).
  • @nhtio/adk/batteries/specialists/ocr — optical character recognition (tesseract.js today).
  • @nhtio/adk/batteries/specialists/caption — image captioning (transformers.js today).
  • @nhtio/adk/batteries/specialists/_shared — the structural contracts and normalization helpers (SpecialistMediaLike, isPcmInput, toBytes, defaultDecodeAudio, etc.) shared across all three modalities.

Unlike @nhtio/adk/batteries/embeddings, which excludes its browser/WebGPU-only WebLLM engine from this level of barrel (reachable only via its own subpath), every specialist engine today is environment-neutral — Node and browser both, no WebGPU requirement — so all three modality barrels are re-exported here in full with nothing held back. Should a browser-only specialist engine be added later, it would be excluded from this barrel for the same reason WebLLM is excluded from the embeddings one, and reachable only through its own subpath.

Each adapter class mirrors the embeddings adapters' shape and lifecycle: constructed with options validated by that engine's own Zod schema, exposing the modality's single-purpose method (transcribe, recognize, describe), and reporting model-load progress through the shared @nhtio/adk/batteries/specialists/ocr!BatteryLifecycleHooks contract.

References ​

BatteryLifecycleBattery ​

Re-exports BatteryLifecycleBattery

BatteryLifecycleCallback ​

Re-exports BatteryLifecycleCallback

BatteryLifecycleHooks ​

Re-exports BatteryLifecycleHooks

BatteryLifecyclePhase ​

Re-exports BatteryLifecyclePhase

BatteryLifecycleReport ​

Re-exports BatteryLifecycleReport

CreateTesseractJsModule ​

Re-exports CreateTesseractJsModule

CreateTesseractJsWorker ​

Re-exports CreateTesseractJsWorker

CreateTransformersJsCaptionPipeline ​

Re-exports CreateTransformersJsCaptionPipeline

CreateTransformersJsSttPipeline ​

Re-exports CreateTransformersJsSttPipeline

DecodeAudioFn ​

Re-exports DecodeAudioFn

defaultDecodeAudio ​

Re-exports defaultDecodeAudio

DescribeOptions ​

Re-exports DescribeOptions

DescribeResult ​

Re-exports DescribeResult

E_INVALID_TESSERACT_JS_OCR_OPTIONS ​

Re-exports E_INVALID_TESSERACT_JS_OCR_OPTIONS

E_INVALID_TRANSFORMERS_JS_CAPTION_OPTIONS ​

Re-exports E_INVALID_TRANSFORMERS_JS_CAPTION_OPTIONS

E_INVALID_TRANSFORMERS_JS_STT_OPTIONS ​

Re-exports E_INVALID_TRANSFORMERS_JS_STT_OPTIONS

E_TESSERACT_JS_OCR_ENGINE_ERROR ​

Re-exports E_TESSERACT_JS_OCR_ENGINE_ERROR

E_TRANSFORMERS_JS_CAPTION_ENGINE_ERROR ​

Re-exports E_TRANSFORMERS_JS_CAPTION_ENGINE_ERROR

E_TRANSFORMERS_JS_STT_ENGINE_ERROR ​

Re-exports E_TRANSFORMERS_JS_STT_ENGINE_ERROR

isPcmInput ​

Re-exports isPcmInput

RecognizeOptions ​

Re-exports RecognizeOptions

RecognizeResult ​

Re-exports RecognizeResult

SpecialistAudioInput ​

Re-exports SpecialistAudioInput

SpecialistBytesInput ​

Re-exports SpecialistBytesInput

SpecialistImageInput ​

Re-exports SpecialistImageInput

SpecialistMediaLike ​

Re-exports SpecialistMediaLike

SpecialistPcmInput ​

Re-exports SpecialistPcmInput

SttSegment ​

Re-exports SttSegment

TesseractJsModule ​

Re-exports TesseractJsModule

TesseractJsOcrAdapter ​

Re-exports TesseractJsOcrAdapter

TesseractJsOcrAdapterOptions ​

Re-exports TesseractJsOcrAdapterOptions

tesseractJsOcrOptionsSchema ​

Re-exports tesseractJsOcrOptionsSchema

TesseractJsWorker ​

Re-exports TesseractJsWorker

toBytes ​

Re-exports toBytes

TranscribeOptions ​

Re-exports TranscribeOptions

TranscribeResult ​

Re-exports TranscribeResult

TransformersJsCaptionAdapter ​

Re-exports TransformersJsCaptionAdapter

TransformersJsCaptionAdapterOptions ​

Re-exports TransformersJsCaptionAdapterOptions

TransformersJsCaptionDataType ​

Re-exports TransformersJsCaptionDataType

TransformersJsCaptionDeviceType ​

Re-exports TransformersJsCaptionDeviceType

TransformersJsCaptionModelSource ​

Re-exports TransformersJsCaptionModelSource

transformersJsCaptionOptionsSchema ​

Re-exports transformersJsCaptionOptionsSchema

TransformersJsCaptionPipeline ​

Re-exports TransformersJsCaptionPipeline

TransformersJsCaptionProgressCallback ​

Re-exports TransformersJsCaptionProgressCallback

TransformersJsSttAdapter ​

Re-exports TransformersJsSttAdapter

TransformersJsSttAdapterOptions ​

Re-exports TransformersJsSttAdapterOptions

transformersJsSttAdapterOptionsSchema ​

Re-exports transformersJsSttAdapterOptionsSchema

TransformersJsSttDataType ​

Re-exports TransformersJsSttDataType

TransformersJsSttDeviceType ​

Re-exports TransformersJsSttDeviceType

TransformersJsSttModelSource ​

Re-exports TransformersJsSttModelSource

TransformersJsSttPipeline ​

Re-exports TransformersJsSttPipeline

TransformersJsSttProgressCallback ​

Re-exports TransformersJsSttProgressCallback

validateTesseractJsOcrOptions ​

Renames and re-exports validateOptions

validateTransformersJsCaptionOptions ​

Renames and re-exports validateOptions

validateTransformersJsSttOptions ​

Re-exports validateTransformersJsSttOptions

Copyright © 2025-present New Horizon Technology LTD