@fugood/bricks-project 2.22.0-beta.15 → 2.22.0-beta.16

package/compile/action-name-map.ts

@@ -610,6 +610,26 @@ export const templateActionNameMap = {
       realtimeVadFreqThold: 'GENERATOR_SPEECH_INFERENCE_REALTIME_VAD_FREQ_THOLD',
     },
   },
+  GENERATOR_VAD_INFERENCE: {
+    GENERATOR_VAD_INFERENCE_DETECT_FILE: {
+      fileUrl: 'GENERATOR_VAD_INFERENCE_FILE_URL',
+      threshold: 'GENERATOR_VAD_INFERENCE_THRESHOLD',
+      minSpeechDurationMs: 'GENERATOR_VAD_INFERENCE_MIN_SPEECH_DURATION_MS',
+      minSilenceDurationMs: 'GENERATOR_VAD_INFERENCE_MIN_SILENCE_DURATION_MS',
+      maxSpeechDurationS: 'GENERATOR_VAD_INFERENCE_MAX_SPEECH_DURATION_S',
+      speechPadMs: 'GENERATOR_VAD_INFERENCE_SPEECH_PAD_MS',
+      samplesOverlap: 'GENERATOR_VAD_INFERENCE_SAMPLES_OVERLAP',
+    },
+    GENERATOR_VAD_INFERENCE_DETECT_DATA: {
+      data: 'GENERATOR_VAD_INFERENCE_DATA',
+      threshold: 'GENERATOR_VAD_INFERENCE_THRESHOLD',
+      minSpeechDurationMs: 'GENERATOR_VAD_INFERENCE_MIN_SPEECH_DURATION_MS',
+      minSilenceDurationMs: 'GENERATOR_VAD_INFERENCE_MIN_SILENCE_DURATION_MS',
+      maxSpeechDurationS: 'GENERATOR_VAD_INFERENCE_MAX_SPEECH_DURATION_S',
+      speechPadMs: 'GENERATOR_VAD_INFERENCE_SPEECH_PAD_MS',
+      samplesOverlap: 'GENERATOR_VAD_INFERENCE_SAMPLES_OVERLAP',
+    },
+  },
   GENERATOR_LLM: {
     GENERATOR_LLM_TOKENIZE: {
       mode: 'GENERATOR_LLM_MODE',
@@ -685,6 +705,12 @@ export const templateActionNameMap = {
       text: 'GENERATOR_GGML_TTS_TEXT',
     },
   },
+  GENERATOR_RERANKER: {
+    GENERATOR_RERANKER_RERANK: {
+      query: 'GENERATOR_RERANKER_QUERY',
+      documents: 'GENERATOR_RERANKER_DOCUMENTS',
+    },
+  },
   GENERATOR_QNN_LLM: {
     GENERATOR_QNN_LLM_GENERATE: {
       prompt: 'GENERATOR_QNN_LLM_PROMPT',

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@fugood/bricks-project",
-  "version": "2.22.0-beta.15",
+  "version": "2.22.0-beta.16",
   "main": "index.ts",
   "scripts": {
     "build": "node scripts/build.js"
@@ -14,5 +14,5 @@
     "lodash": "^4.17.4",
     "uuid": "^8.3.1"
   },
-  "gitHead": "4995b78120bd6e75df65a302398a9ba3070da3e7"
+  "gitHead": "d8231fa6787b8d95bc2bec94a86ceba596cf57f0"
 }

package/tools/postinstall.ts

@@ -1,15 +1,7 @@
 import { $ } from 'bun'
 import { stat, readFile, writeFile } from 'fs/promises'
-const cwd = process.cwd()
-
-const libFiles = ['types', 'utils', 'index.ts']
 
-await $`mkdir -p ${cwd}/project`
-for (const file of libFiles) {
-  await $`cp -r ${__dirname}/../${file} ${cwd}/project`
-}
-
-console.log('Copied files to project/')
+const cwd = process.cwd()
 
 async function exists(f: string) {
   try {
@@ -20,6 +12,21 @@ async function exists(f: string) {
   }
 }
 
+// handle flag --skip-copy
+const skipCopyProject = process.argv.includes('--skip-copy-project')
+if (skipCopyProject) {
+  console.log('Skipping copy of files to project/')
+} else {
+
+  const libFiles = ['types', 'utils', 'index.ts']
+  
+  await $`mkdir -p ${cwd}/project`
+  for (const file of libFiles) {
+    await $`cp -r ${__dirname}/../${file} ${cwd}/project`
+  }
+  console.log('Copied files to project/')
+}
+
 const projectMcpServer = {
   command: 'bun',
   args: [`${cwd}/node_modules/@fugood/bricks-project/tools/mcp-server.ts`],

package/types/common.ts

@@ -65,7 +65,7 @@ export type ItemBrickID = string
 
 export type EventAction = {
   handler: 'system' | (() => Brick | Generator) | SubspaceID | ItemBrickID
-  action: Action
+  action: ActionWithParams | ActionWithDataParams
   waitAsync?: boolean
 }
 

package/types/generators.ts

@@ -4856,6 +4856,7 @@ Default property:
   "doSample": true,
   "outputType": "play",
   "cacheGenerated": true,
+  "speed": 1,
   "autoInferEnable": false,
   "softBreakRegex": "^[^\\r\\n\\t\\f\\v]*([\\r\\n]+|[。！？!?.]\\B)",
   "hardBreakTime": 500,
@@ -4872,6 +4873,8 @@ Default property:
       | 'BricksDisplay/vits-eng'
       | 'BricksDisplay/vits-cmn'
       | 'BricksDisplay/ellie-Bert-VITS2'
+      | 'onnx-community/Kokoro-82M-v1.0-ONNX'
+      | 'onnx-community/Kokoro-82M-v1.1-zh-ONNX'
       | 'onnx-community/OuteTTS-1.0-0.6B-ONNX'
       | 'mms-tts-ara (NC)'
       | 'mms-tts-deu (NC)'
@@ -4888,7 +4891,14 @@ Default property:
       | 'speecht5_tts'
       | DataLink
     /* Model type */
-    modelType?: 'auto' | 'vits' | 'bert_vits2' | 'speecht5' | 'outetts-1.0' | DataLink
+    modelType?:
+      | 'auto'
+      | 'vits'
+      | 'bert_vits2'
+      | 'style_text_to_speech_2'
+      | 'speecht5'
+      | 'outetts-1.0'
+      | DataLink
     /* Load quantized model (deprecated, use `quantizeType` instead) */
     quantized?: boolean | DataLink
     /* Quantize type */
@@ -4911,7 +4921,7 @@ Default property:
     /* Custom vocoder model  
  Choose model from https://huggingface.co/models?library=transformers.js&other=hifigan */
     customVocoderModel?: string | DataLink
-    /* XVector speaker embedding for HiFi-GAN */
+    /* Speaker embedding, for SpeechT5 or StyleTTS (Kokoro) */
     speakerEmbedUrl?: string | DataLink
     /* MD5 checksum of `speakerEmbedUrl` */
     speakerEmbedMd5?: string | DataLink
@@ -4929,6 +4939,8 @@ Default property:
     outputType?: 'play' | 'file' | DataLink
     /* Enable cache for generated audio */
     cacheGenerated?: boolean | DataLink
+    /* Speed of the generated audio, for StyleTTS (Kokoro) */
+    speed?: number | DataLink
     /* Text to generate */
     prompt?: string | DataLink
     /* Auto inference when prompt changes */
@@ -5909,6 +5921,207 @@ export type GeneratorSpeechInference = Generator &
     >
   }
 
+/* Load the model */
+export type GeneratorVadInferenceActionLoadModel = Action & {
+  __actionName: 'GENERATOR_VAD_INFERENCE_LOAD_MODEL'
+}
+
+/* Detect speech in audio file. You can provide `File URL` property, if not provided, it will use the default `File URL` */
+export type GeneratorVadInferenceActionDetectFile = ActionWithParams & {
+  __actionName: 'GENERATOR_VAD_INFERENCE_DETECT_FILE'
+  params?: Array<
+    | {
+        input: 'fileUrl'
+        value?: string | DataLink | EventProperty
+        mapping?: string
+      }
+    | {
+        input: 'threshold'
+        value?: number | DataLink | EventProperty
+        mapping?: string
+      }
+    | {
+        input: 'minSpeechDurationMs'
+        value?: number | DataLink | EventProperty
+        mapping?: string
+      }
+    | {
+        input: 'minSilenceDurationMs'
+        value?: number | DataLink | EventProperty
+        mapping?: string
+      }
+    | {
+        input: 'maxSpeechDurationS'
+        value?: number | DataLink | EventProperty
+        mapping?: string
+      }
+    | {
+        input: 'speechPadMs'
+        value?: number | DataLink | EventProperty
+        mapping?: string
+      }
+    | {
+        input: 'samplesOverlap'
+        value?: number | DataLink | EventProperty
+        mapping?: string
+      }
+  >
+}
+
+/* Detect speech in audio data. Currently only support base64 encoded audio data (16-bit PCM, mono, 16kHz) */
+export type GeneratorVadInferenceActionDetectData = ActionWithParams & {
+  __actionName: 'GENERATOR_VAD_INFERENCE_DETECT_DATA'
+  params?: Array<
+    | {
+        input: 'data'
+        value?: any | EventProperty
+        mapping?: string
+      }
+    | {
+        input: 'threshold'
+        value?: number | DataLink | EventProperty
+        mapping?: string
+      }
+    | {
+        input: 'minSpeechDurationMs'
+        value?: number | DataLink | EventProperty
+        mapping?: string
+      }
+    | {
+        input: 'minSilenceDurationMs'
+        value?: number | DataLink | EventProperty
+        mapping?: string
+      }
+    | {
+        input: 'maxSpeechDurationS'
+        value?: number | DataLink | EventProperty
+        mapping?: string
+      }
+    | {
+        input: 'speechPadMs'
+        value?: number | DataLink | EventProperty
+        mapping?: string
+      }
+    | {
+        input: 'samplesOverlap'
+        value?: number | DataLink | EventProperty
+        mapping?: string
+      }
+  >
+}
+
+/* Clear downloaded files (model, audio) & current jobs */
+export type GeneratorVadInferenceActionClearDownload = Action & {
+  __actionName: 'GENERATOR_VAD_INFERENCE_CLEAR_DOWNLOAD'
+}
+
+/* Release context */
+export type GeneratorVadInferenceActionReleaseContext = Action & {
+  __actionName: 'GENERATOR_VAD_INFERENCE_RELEASE_CONTEXT'
+}
+
+interface GeneratorVadInferenceDef {
+  /*
+Default property:
+{
+  "init": false,
+  "modelName": "silero-v5.1.2",
+  "modelUseGPU": true,
+  "modelThreads": 4,
+  "detectThreshold": 0.5,
+  "detectMinSpeechDurationMs": 250,
+  "detectMinSilenceDurationMs": 100,
+  "detectMaxSpeechDurationS": 30,
+  "detectSpeechPadMs": 30,
+  "detectSamplesOverlap": 0.1
+}
+  */
+  property?: {
+    /* Initialize the VAD context on generator initialization  
+ Please note that it will take some RAM depending on the model size */
+    init?: boolean | DataLink
+    /* Use model name, currently only supports the Silero VAD model.  
+ The model download progress will be done in preload stage or  the generator initialization stage.  
+ You can also choose `custom` option and set `Model URL` and `Model SHA1` to use your own model */
+    modelName?: 'custom' | 'silero-v5.1.2' | DataLink
+    /* The URL or path of model  
+ We used `ggml` format model, please refer to https://huggingface.co/ggml-org/whisper-vad */
+    modelUrl?: string | DataLink
+    /* Hash type of model */
+    modelHashType?: 'md5' | 'sha256' | 'sha1' | DataLink
+    /* Hash of model */
+    modelHash?: string | DataLink
+    /* Use GPU Acceleration for inference. Currently iOS only. */
+    modelUseGPU?: boolean | DataLink
+    /* Number of threads to use for processing */
+    modelThreads?: number | DataLink
+    /* Speech probability threshold (0.0-1.0) */
+    detectThreshold?: number | DataLink
+    /* Minimum speech duration in milliseconds */
+    detectMinSpeechDurationMs?: number | DataLink
+    /* Minimum silence duration in milliseconds */
+    detectMinSilenceDurationMs?: number | DataLink
+    /* Maximum speech duration in seconds */
+    detectMaxSpeechDurationS?: number | DataLink
+    /* Padding around speech segments in milliseconds */
+    detectSpeechPadMs?: number | DataLink
+    /* Overlap between analysis windows (0.0-1.0) */
+    detectSamplesOverlap?: number | DataLink
+    /* The file URL or path to be analyzed.  
+ It only supported `wav` format with 16kHz sample rate & single (mono) channel */
+    detectFileUrl?: string | DataLink
+    /* MD5 of file to be analyzed */
+    detectFileMd5?: string | DataLink
+  }
+  events?: {
+    /* Event triggered when load is done */
+    onContextStateChange?: Array<EventAction>
+    /* Event triggered when error occurs */
+    onError?: Array<EventAction>
+    /* Event triggered when got detection result */
+    onDetected?: Array<EventAction>
+  }
+  outlets?: {
+    /* Context state */
+    contextState?: () => Data
+    /* Context details */
+    contextDetails?: () => Data
+    /* Is detecting */
+    isDetecting?: () => Data
+    /* Detection segments result */
+    detectionSegments?: () => Data
+    /* Detection details */
+    detectionDetails?: () => Data
+  }
+}
+
+/* Local Voice Activity Detection (VAD) inference based on GGML and [whisper.rn](https://github.com/mybigday/whisper.rn) */
+export type GeneratorVadInference = Generator &
+  GeneratorVadInferenceDef & {
+    templateKey: 'GENERATOR_VAD_INFERENCE'
+    switches: Array<
+      SwitchDef &
+        GeneratorVadInferenceDef & {
+          conds?: Array<{
+            method: '==' | '!=' | '>' | '<' | '>=' | '<='
+            cond:
+              | SwitchCondInnerStateCurrentCanvas
+              | SwitchCondData
+              | {
+                  __typename: 'SwitchCondInnerStateOutlet'
+                  outlet:
+                    | 'contextState'
+                    | 'contextDetails'
+                    | 'isDetecting'
+                    | 'detectionSegments'
+                    | 'detectionDetails'
+                  value: any
+                }
+          }>
+        }
+    >
+  }
+
 /* Load the model */
 export type GeneratorLLMActionLoadModel = Action & {
   __actionName: 'GENERATOR_LLM_LOAD_MODEL'
@@ -6731,6 +6944,149 @@ export type GeneratorGGMLTTS = Generator &
     >
   }
 
+/* Load the model */
+export type GeneratorRerankerActionLoadModel = Action & {
+  __actionName: 'GENERATOR_RERANKER_LOAD_MODEL'
+}
+
+/* Rerank documents based on query relevance */
+export type GeneratorRerankerActionRerank = ActionWithParams & {
+  __actionName: 'GENERATOR_RERANKER_RERANK'
+  params?: Array<
+    | {
+        input: 'query'
+        value?: string | DataLink | EventProperty
+        mapping?: string
+      }
+    | {
+        input: 'documents'
+        value?: Array<any> | DataLink | EventProperty
+        mapping?: string
+      }
+  >
+}
+
+/* Release context */
+export type GeneratorRerankerActionReleaseContext = Action & {
+  __actionName: 'GENERATOR_RERANKER_RELEASE_CONTEXT'
+}
+
+interface GeneratorRerankerDef {
+  /*
+Default property:
+{
+  "init": false,
+  "contextSize": 512,
+  "batchSize": 512,
+  "uBatchSize": 512,
+  "accelVariant": "default",
+  "mainGpu": 0,
+  "gpuLayers": 0,
+  "useMlock": true,
+  "useMmap": true,
+  "normalize": 1
+}
+  */
+  property?: {
+    /* Initialize the Reranker context on generator initialization */
+    init?: boolean | DataLink
+    /* The URL or path of reranker model (GGUF format) */
+    modelUrl?: string | DataLink
+    /* Hash of reranker model */
+    modelHash?: string | DataLink
+    /* Hash type of reranker model */
+    modelHashType?: 'md5' | 'sha256' | 'sha1' | DataLink
+    /* Context size (0 ~ 4096) (Default to 512) */
+    contextSize?: number | DataLink
+    /* Logical batch size for processing (default: 512) */
+    batchSize?: number | DataLink
+    /* Physical maximum batch size (default: 512) */
+    uBatchSize?: number | DataLink
+    /* Accelerator variant (default: default) */
+    accelVariant?:
+      | 'default'
+      | 'avx'
+      | 'avx2'
+      | 'avx512'
+      | 'metal'
+      | 'opencl'
+      | 'vulkan'
+      | 'cuda'
+      | 'rocm'
+      | DataLink
+    /* Main GPU index (default: 0) */
+    mainGpu?: number | DataLink
+    /* Number of layers to store in VRAM (default: 0) */
+    gpuLayers?: number | DataLink
+    /* Maximum number of threads to use (default: auto) */
+    maxThreads?: number | DataLink
+    /* Use mlock to keep model in memory (default: true) */
+    useMlock?: boolean | DataLink
+    /* Use mmap for model loading (default: true) */
+    useMmap?: boolean | DataLink
+    /* Query text for reranking */
+    query?: string | DataLink
+    /* Array of documents to rerank */
+    documents?: Array<string | DataLink> | DataLink
+    /* Normalize reranking scores (default: from model config) */
+    normalize?: number | DataLink | boolean | DataLink | DataLink
+    /* Maximum number of documents to return (default: unlimited) */
+    topK?: number | DataLink
+  }
+  events?: {
+    /* Event triggered when the reranker context state changes (loading, ready, error, released) */
+    onContextStateChange?: Array<EventAction>
+    /* Event triggered when an error occurs during reranker operations */
+    onError?: Array<EventAction>
+  }
+  outlets?: {
+    /* Current state of the reranker context (loading, ready, error, released) */
+    contextState?: () => Data
+    /* Loading progress of the reranker model (0-100) */
+    contextLoadProgress?: () => Data
+    /* Detailed information about the reranker context including instance ID and processing status */
+    contextDetails?: () => Data
+    /* Result of the reranking operation containing scored and ranked documents */
+    rerankResult?: () => Data
+    /* Boolean indicating whether the reranker is currently processing a request */
+    isProcessing?: () => Data
+  }
+}
+
+/* Local rerank based on GGML and [llama.cpp](https://github.com/ggerganov/llama.cpp)  
+
+ ## Notice  
+ - The device RAM must be larger than 8GB  
+ - iOS: Supported GPU acceleration, recommended use M1+ / A17+ chip device  
+ - macOS: Supported GPU acceleration, recommended use M1+ chip device  
+ - Android: Currently not supported GPU acceleration (Coming soon), recommended use Android 13+ system  
+ - Linux / Windows: Supported GPU acceleration, currently only Vulkan backend available */
+export type GeneratorReranker = Generator &
+  GeneratorRerankerDef & {
+    templateKey: 'GENERATOR_RERANKER'
+    switches: Array<
+      SwitchDef &
+        GeneratorRerankerDef & {
+          conds?: Array<{
+            method: '==' | '!=' | '>' | '<' | '>=' | '<='
+            cond:
+              | SwitchCondInnerStateCurrentCanvas
+              | SwitchCondData
+              | {
+                  __typename: 'SwitchCondInnerStateOutlet'
+                  outlet:
+                    | 'contextState'
+                    | 'contextLoadProgress'
+                    | 'contextDetails'
+                    | 'rerankResult'
+                    | 'isProcessing'
+                  value: any
+                }
+          }>
+        }
+    >
+  }
+
 /* Load the model */
 export type GeneratorQnnLlmActionLoadModel = Action & {
   __actionName: 'GENERATOR_QNN_LLM_LOAD_MODEL'

package/utils/event-props.ts

@@ -761,6 +761,19 @@ export const templateEventPropsMap = {
       'GENERATOR_SPEECH_INFERENCE_TRANSCRIBE_TIME', // type: number
     ],
   },
+  GENERATOR_VAD_INFERENCE: {
+    onContextStateChange: [
+      'GENERATOR_VAD_INFERENCE_CONTEXT_STATE', // type: string
+      'GENERATOR_VAD_INFERENCE_CONTEXT_DETAILS', // type: object
+    ],
+    onError: [
+      'GENERATOR_VAD_INFERENCE_ERROR', // type: string
+    ],
+    onDetected: [
+      'GENERATOR_VAD_INFERENCE_DETECTION_SEGMENTS', // type: array
+      'GENERATOR_VAD_INFERENCE_DETECTION_TIME', // type: number
+    ],
+  },
   GENERATOR_LLM: {
     onContextStateChange: [
       'GENERATOR_LLM_CONTEXT_STATE', // type: string
@@ -799,6 +812,7 @@ export const templateEventPropsMap = {
       'GENERATOR_GGML_TTS_ERROR', // type: string
     ],
   },
+  GENERATOR_RERANKER: {},
   GENERATOR_QNN_LLM: {
     onContextStateChange: [
       'GENERATOR_QNN_LLM_CONTEXT_STATE', // type: string