@dxos/plugin-transcription 0.8.3 → 0.8.4-main.1068cf700f

package/src/segments-normalization/normalization.test.ts

@@ -2,25 +2,22 @@
 // Copyright 2025 DXOS.org
 //
 
-import { effect } from '@preact/signals-core';
 import { describe, test } from 'vitest';
 
-import { EdgeAiServiceClient, OllamaAiServiceClient } from '@dxos/ai';
-import { AI_SERVICE_ENDPOINT } from '@dxos/ai/testing';
 import { scheduleTaskInterval } from '@dxos/async';
 import { Context } from '@dxos/context';
 import { Obj } from '@dxos/echo';
+import { createQueueDXN } from '@dxos/echo/internal';
 import { MemoryQueue } from '@dxos/echo-db';
-import { createQueueDXN } from '@dxos/echo-schema';
-import { FunctionExecutor, ServiceContainer } from '@dxos/functions';
+import { FunctionExecutor, ServiceContainer } from '@dxos/functions-runtime';
 import { log } from '@dxos/log';
-import { DataType } from '@dxos/schema';
+import { type Actor, Message } from '@dxos/types';
 
 import { MessageNormalizer } from './message-normalizer';
 import { type MessageWithRangeId, sentenceNormalization } from './normalization';
 import { getActorId } from './utils';
 
-const sender: DataType.Actor = {
+const sender: Actor.Actor = {
   identityDid: 'did:key:123',
 };
 
@@ -50,34 +47,33 @@ const messages: MessageWithRangeId[] = [
   // No punctuation.
   'in classical physics objects have well-defined properties such as position speed and momentum',
 ].map((string, index) =>
-  Obj.make(DataType.Message, {
-    created: new Date(Date.now() + 1000 * index).toISOString(),
+  Obj.make(Message.Message, {
+    created: new Date(Date.now() + 1_000 * index).toISOString(),
     sender,
-    blocks: [{ type: 'transcription', started: new Date(Date.now() + 1000 * index).toISOString(), text: string }],
-    rangeId: [],
-  } as any),
+    blocks: [{ _tag: 'transcript', started: new Date(Date.now() + 1_000 * index).toISOString(), text: string }],
+  }),
 );
 
-const REMOTE_AI = true;
+// const REMOTE_AI = true;
 
 describe.skip('SentenceNormalization', () => {
   const getExecutor = () => {
     return new FunctionExecutor(
       new ServiceContainer().setServices({
-        ai: {
-          client: REMOTE_AI
-            ? new EdgeAiServiceClient({
-                endpoint: AI_SERVICE_ENDPOINT.REMOTE,
-                defaultGenerationOptions: {
-                  model: '@anthropic/claude-3-5-sonnet-20241022',
-                },
-              })
-            : new OllamaAiServiceClient({
-                overrides: {
-                  model: 'llama3.1:8b',
-                },
-              }),
-        },
+        // ai: {
+        //   client: REMOTE_AI
+        //     ? new Edge AiServiceClient({
+        //         endpoint: AI_SERVICE_ENDPOINT.REMOTE,
+        //         defaultGenerationOptions: {
+        //           model: '@anthropic/claude-3-5-sonnet-20241022',
+        //         },
+        //       })
+        //     : new Ollama AiServiceClient({
+        //         overrides: {
+        //           model: 'llama3.1:8b',
+        //         },
+        //       }),
+        // },
       }),
     );
   };
@@ -100,6 +96,7 @@ describe.skip('SentenceNormalization', () => {
         buffer = sentences.slice(activeSentenceIndex);
       }
     }
+
     log.info('sentences', {
       originalMessages: JSON.stringify(messages, null, 2),
       sentences: JSON.stringify(sentences, null, 2),
@@ -110,7 +107,7 @@ describe.skip('SentenceNormalization', () => {
   test.only('queue', { timeout: 120_000 }, async () => {
     // Create queue.
     // TODO(dmaretskyi): Use space.queues.create() instead.
-    const queue = new MemoryQueue<DataType.Message>(createQueueDXN());
+    const queue = new MemoryQueue<Message.Message>(createQueueDXN());
     const ctx = new Context();
     let idx = 0;
     scheduleTaskInterval(
@@ -135,10 +132,6 @@ describe.skip('SentenceNormalization', () => {
 
     // Start normalizer.
     await normalizer.open();
-    effect(() => {
-      log.info('normalizer');
-      log.info(JSON.stringify(queue.objects, null, 2));
-    });
 
     await new Promise(() => {});
   });

package/src/segments-normalization/normalization.ts

@@ -2,20 +2,24 @@
 // Copyright 2025 DXOS.org
 //
 
-import { Schema } from 'effect';
+// ISSUE(burdon): defineFunction
+// @ts-nocheck
 
-import { DEFAULT_EDGE_MODEL, Message } from '@dxos/ai';
-import { AISession } from '@dxos/assistant';
+import * as Schema from 'effect/Schema';
+
+import { AiService, DEFAULT_EDGE_MODEL } from '@dxos/ai';
+import { AiSession } from '@dxos/assistant';
 import { Obj } from '@dxos/echo';
-import { AiService, defineFunction } from '@dxos/functions';
+import { defineFunction } from '@dxos/functions';
 import { ObjectId } from '@dxos/keys';
 import { log } from '@dxos/log';
-import { DataType } from '@dxos/schema';
+import { Message } from '@dxos/types';
+import { trim } from '@dxos/util';
 
 const MessageWithRangeId = Schema.extend(
-  DataType.Message,
+  Message.Message,
   Schema.Struct({
-    // TODO(mykola): Move to meta in DataType.Message. Use `DataType.Message` instead.
+    // TODO(mykola): Move to meta in Message. Use `Message.Message` instead.
     rangeId: Schema.optional(Schema.Array(Schema.String)).annotations({
       description: 'The IDs of the messages that contain the sentences.',
     }),
@@ -25,7 +29,7 @@ const MessageWithRangeId = Schema.extend(
 export interface MessageWithRangeId extends Schema.Schema.Type<typeof MessageWithRangeId> {}
 
 export const NormalizationInput = Schema.Struct({
-  // TODO(mykola): Move to meta in DataType.Message. Use `DataType.Message` instead.
+  // TODO(mykola): Move to meta in Message. Use `Message.Message` instead.
   messages: Schema.Array(MessageWithRangeId).annotations({
     description: 'Messages to normalize into sentences.',
   }),
@@ -39,40 +43,6 @@ export const NormalizationOutput = Schema.Struct({
 });
 export interface NormalizationOutput extends Schema.Schema.Type<typeof NormalizationOutput> {}
 
-const prompt = `
-You are observing a real-time transcript of a single person speaking.
-The transcription is delivered in chunks of 10 seconds or less. As a result, individual sentences may be split across multiple messages, or multiple sentences may appear within a single message. Additionally, because this is real-time transcription, punctuation and capitalization may be incorrect or missing.
-
-# Task Description:
-  - Your task is to detect and reconstruct broken or incomplete sentences by merging segments into coherent, grammatically correct sentences where appropriate.
-
-# Input Format:
-  - The input is an array of messages that contain transcription blocks.
-  - Each message is a DataType.Message, as described in the output format.
-
-# Output Format:
-  - You have been provided with the tool that defines the output format; make sure to query it.
-  - Do not output anything other than the expected format.
-
-# Segment Handling Rules:
-  - Sort messages by timestamp.
-  - Leave ID of the first message in a sentence.
-  - Each divided sentence should be added to the output as a separate message or block within existing messages, preserving the original order.
-  - If one message contains multiple sentences, you should not split them.
-  - The last sentence may be incomplete; it should still be output as a separate block or message.
-  - Keep track of the original timestamps of the messages and blocks and, use the timestamp of the first message/block of a sentence as the 'started' field of the merged message.
-  - The 'rangeId' field of the merged message should include 'messageId'-s and 'rangeId'-s of all messages that have been used to construct this message.
-
-# Punctuation and Capitalization:
-  - Do not rely on the original punctuation and capitalization—they may be incorrect.
-  - Use logical reasoning to apply appropriate punctuation (e.g., period, comma, question mark, exclamation mark) and capitalization.
-
-# Restrictions:
-  - Do not alter the order of the original messages; maintain the natural flow of speech.
-  - Do not interpret or infer meaning beyond reconstructing sentences.
-  - Do not add or remove any words or phrases.
-`;
-
 export const sentenceNormalization = defineFunction<NormalizationInput, NormalizationOutput>({
   description: 'Post process of transcription for sentence normalization',
   inputSchema: NormalizationInput,
@@ -80,21 +50,27 @@ export const sentenceNormalization = defineFunction<NormalizationInput, Normaliz
   handler: async ({ data: { messages }, context }) => {
     log.info('input', { messages });
     const ai = context.getService(AiService);
-    const session = new AISession({ operationModel: 'configured' });
+    const session = new AiSession({ operationModel: 'configured' });
 
-    const response = await session.runStructured(NormalizationOutput, {
-      generationOptions: { model: DEFAULT_EDGE_MODEL },
+    // TODO(dmaretskyi): This got broken after effect-ai transition.
+    const response = session.runStructured(NormalizationOutput, {
+      generationOptions: {
+        model: DEFAULT_EDGE_MODEL,
+      },
       client: ai.client,
       tools: [],
       artifacts: [],
       history: [
-        Obj.make(Message, {
-          role: 'user',
-          content: messages.map((message) => ({ type: 'text', text: JSON.stringify(message) }) as const),
+        Obj.make(Message.Message, {
+          created: new Date().toISOString(),
+          sender: {
+            role: 'user',
+          },
+          blocks: messages.map((message) => ({ _tag: 'text', text: JSON.stringify(message) }) as const),
         }),
       ],
       prompt,
-    });
+    }) as any;
 
     response.sentences.forEach((sentence) => {
       sentence.id = ObjectId.random();
@@ -103,3 +79,37 @@ export const sentenceNormalization = defineFunction<NormalizationInput, Normaliz
     return response;
   },
 });
+
+const prompt = trim`
+  You are observing a real-time transcript of a single person speaking.
+  The transcription is delivered in chunks of 10 seconds or less. As a result, individual sentences may be split across multiple messages, or multiple sentences may appear within a single message. Additionally, because this is real-time transcription, punctuation and capitalization may be incorrect or missing.
+
+  # Task Description:
+    - Your task is to detect and reconstruct broken or incomplete sentences by merging segments into coherent, grammatically correct sentences where appropriate.
+
+  # Input Format:
+    - The input is an array of messages that contain transcription blocks.
+    - Each message is a Message.Message, as described in the output format.
+
+  # Output Format:
+    - You have been provided with the tool that defines the output format; make sure to query it.
+    - Do not output anything other than the expected format.
+
+  # Segment Handling Rules:
+    - Sort messages by timestamp.
+    - Leave ID of the first message in a sentence.
+    - Each divided sentence should be added to the output as a separate message or block within existing messages, preserving the original order.
+    - If one message contains multiple sentences, you should not split them.
+    - The last sentence may be incomplete; it should still be output as a separate block or message.
+    - Keep track of the original timestamps of the messages and blocks and, use the timestamp of the first message/block of a sentence as the 'started' field of the merged message.
+    - The 'rangeId' field of the merged message should include 'messageId'-s and 'rangeId'-s of all messages that have been used to construct this message.
+
+  # Punctuation and Capitalization:
+    - Do not rely on the original punctuation and capitalization—they may be incorrect.
+    - Use logical reasoning to apply appropriate punctuation (e.g., period, comma, question mark, exclamation mark) and capitalization.
+
+  # Restrictions:
+    - Do not alter the order of the original messages; maintain the natural flow of speech.
+    - Do not interpret or infer meaning beyond reconstructing sentences.
+    - Do not add or remove any words or phrases.
+`;

package/src/segments-normalization/utils.ts

@@ -2,8 +2,8 @@
 // Copyright 2025 DXOS.org
 //
 
-import { type DataType } from '@dxos/schema';
+import { type Actor } from '@dxos/types';
 
-export const getActorId = (actor: DataType.Actor) => {
+export const getActorId = (actor: Actor.Actor) => {
   return actor.identityDid || actor.identityKey || actor.email || actor.name || '';
 };

package/src/testing/testing.ts

@@ -2,24 +2,23 @@
 // Copyright 2025 DXOS.org
 //
 
-import { Schema } from 'effect';
+import * as Schema from 'effect/Schema';
 import { useEffect, useMemo, useState } from 'react';
 
-import { EdgeAiServiceClient } from '@dxos/ai';
-import { AI_SERVICE_ENDPOINT } from '@dxos/ai/testing';
-import { extractionAnthropicFn, processTranscriptMessage } from '@dxos/assistant';
+import { extractionAnthropicFunction, processTranscriptMessage } from '@dxos/assistant/extraction';
 import { scheduleTaskInterval } from '@dxos/async';
 import { Filter, type Queue } from '@dxos/client/echo';
 import { Context } from '@dxos/context';
 import { type Key, Obj, Ref, Type } from '@dxos/echo';
-import { createQueueDXN } from '@dxos/echo-schema';
-import { FunctionExecutor, ServiceContainer } from '@dxos/functions';
+import { createQueueDXN } from '@dxos/echo/internal';
+import { FunctionExecutor, ServiceContainer } from '@dxos/functions-runtime';
 import { IdentityDid } from '@dxos/keys';
 import { log } from '@dxos/log';
 import { faker } from '@dxos/random';
-import { useQueue, type Space } from '@dxos/react-client/echo';
-import { DataType } from '@dxos/schema';
-import { Testing, seedTestData } from '@dxos/schema/testing';
+import { type Space, useQueue } from '@dxos/react-client/echo';
+import { TestSchema } from '@dxos/schema/testing';
+import { type ContentBlock, Message, Organization, Person } from '@dxos/types';
+import { seedTestData } from '@dxos/types/testing';
 
 // TODO(burdon): Reconcile with plugin-markdown. Move to @dxos/schema/testing.
 export const TestItem = Schema.Struct({
@@ -32,7 +31,7 @@ export const TestItem = Schema.Struct({
     description: 'Product description',
   }),
 }).pipe(
-  Type.Obj({
+  Type.object({
     typename: 'dxos.org/type/Test',
     version: '0.1.0',
   }),
@@ -40,7 +39,7 @@ export const TestItem = Schema.Struct({
 
 // TODO(wittjosiah): Make builder generic and reuse for all message types.
 abstract class AbstractMessageBuilder {
-  abstract createMessage(numSegments?: number): Promise<DataType.Message>;
+  abstract createMessage(numSegments?: number): Promise<Message.Message>;
 }
 
 /**
@@ -61,27 +60,32 @@ export class MessageBuilder extends AbstractMessageBuilder {
     super();
   }
 
-  override async createMessage(numSegments = 1): Promise<DataType.Message> {
-    return Obj.make(DataType.Message, {
+  override async createMessage(numSegments = 1): Promise<Message.Message> {
+    return Obj.make(Message.Message, {
       created: this.next().toISOString(),
       sender: faker.helpers.arrayElement(this.users),
       blocks: Array.from({ length: numSegments }).map(() => this.createBlock()),
     });
   }
 
-  createBlock(): DataType.MessageBlock.Transcription {
+  createBlock(): ContentBlock.Transcript {
     let text = faker.lorem.paragraph();
     if (this._space) {
       const label = faker.commerce.productName();
-      const obj = this._space.db.add(Obj.make(TestItem, { title: label, description: faker.lorem.paragraph() }));
+      const obj = this._space.db.add(
+        Obj.make(TestItem, {
+          title: label,
+          description: faker.lorem.paragraph(),
+        }),
+      );
       const dxn = Ref.make(obj).dxn.toString();
       const words = text.split(' ');
-      words.splice(Math.floor(Math.random() * words.length), 0, `[${label}][${dxn}]`);
+      words.splice(Math.floor(Math.random() * words.length), 0, `[${label}](${dxn})`);
       text = words.join(' ');
     }
 
     return {
-      type: 'transcription',
+      _tag: 'transcript',
       started: this.next().toISOString(),
       text,
     };
@@ -95,15 +99,17 @@ export class MessageBuilder extends AbstractMessageBuilder {
 
 // TODO(burdon): Reconcile with BlockBuilder.
 class EntityExtractionMessageBuilder extends AbstractMessageBuilder {
-  AiService = new EdgeAiServiceClient({
-    endpoint: AI_SERVICE_ENDPOINT.REMOTE,
-  });
-
-  executor = new FunctionExecutor(new ServiceContainer().setServices({ ai: { client: this.AiService } }));
+  executor = new FunctionExecutor(
+    new ServiceContainer().setServices({
+      // ai: {
+      //   client: this.AiService,
+      // },
+    }),
+  );
 
   space: Space | undefined;
   currentMessage: number = 0;
-  transcriptMessages: DataType.Message[] = [];
+  transcriptMessages: Message.Message[] = [];
 
   async connect(space: Space): Promise<void> {
     this.space = space;
@@ -111,14 +117,18 @@ class EntityExtractionMessageBuilder extends AbstractMessageBuilder {
     this.transcriptMessages = transcriptMessages;
   }
 
-  override async createMessage(): Promise<DataType.Message> {
+  override async createMessage(): Promise<Message.Message> {
     if (!this.space) {
       throw new Error('Space not connected');
     }
 
-    const { objects } = await this.space.db
+    const objects = await this.space.db
       .query(
-        Filter.or(Filter.type(DataType.Person), Filter.type(DataType.Organization), Filter.type(Testing.DocumentType)),
+        Filter.or(
+          Filter.type(Person.Person),
+          Filter.type(Organization.Organization),
+          Filter.type(TestSchema.DocumentType),
+        ),
       )
       .run();
 
@@ -130,7 +140,7 @@ class EntityExtractionMessageBuilder extends AbstractMessageBuilder {
     const { message: enhancedMessage } = await processTranscriptMessage({
       input: { message },
       executor: this.executor,
-      function: extractionAnthropicFn,
+      function: extractionAnthropicFunction,
     });
 
     return enhancedMessage;
@@ -142,7 +152,7 @@ type UseTestTranscriptionQueue = (
   queueId?: Key.ObjectId,
   running?: boolean,
   interval?: number,
-) => Queue<DataType.Message> | undefined;
+) => Queue<Message.Message> | undefined;
 
 /**
  * Test transcriptionqueue.
@@ -155,7 +165,7 @@ export const useTestTranscriptionQueue: UseTestTranscriptionQueue = (
 ) => {
   // TODO(dmaretskyi): Use space.queues.create() instead.
   const queueDxn = useMemo(() => (space ? createQueueDXN(space.id, queueId) : undefined), [space, queueId]);
-  const queue = useQueue<DataType.Message>(queueDxn);
+  const queue = useQueue<Message.Message>(queueDxn);
   const builder = useMemo(() => new MessageBuilder(space), [space]);
 
   useEffect(() => {
@@ -165,7 +175,7 @@ export const useTestTranscriptionQueue: UseTestTranscriptionQueue = (
 
     const i = setInterval(() => {
       void builder.createMessage(Math.ceil(Math.random() * 3)).then(async (message) => {
-        await queue.append([Obj.make(DataType.Message, message)]);
+        await queue.append([Obj.make(Message.Message, message)]);
       });
     }, interval);
     return () => clearInterval(i);
@@ -186,7 +196,7 @@ export const useTestTranscriptionQueueWithEntityExtraction: UseTestTranscription
 ) => {
   // TODO(dmaretskyi): Use space.queues.create() instead.
   const queueDxn = useMemo(() => (space ? createQueueDXN(space.id, queueId) : undefined), [space, queueId]);
-  const queue = useQueue<DataType.Message>(queueDxn);
+  const queue = useQueue<Message.Message>(queueDxn);
   const [builder] = useState(() => new EntityExtractionMessageBuilder());
 
   useEffect(() => {
@@ -203,7 +213,7 @@ export const useTestTranscriptionQueueWithEntityExtraction: UseTestTranscription
       ctx,
       async () => {
         const message = await builder.createMessage();
-        void queue.append([Obj.make(DataType.Message, message)]);
+        void queue.append([Obj.make(Message.Message, message)]);
       },
       interval,
     );

package/src/transcriber/index.ts

@@ -2,7 +2,7 @@
 // Copyright 2025 DXOS.org
 //
 
-export * from './audio-recorder';
+export type * from './audio-recorder';
 export * from './media-stream-recorder';
 export * from './transcriber';
 export * from './transcription-manager';

package/src/transcriber/media-stream-recorder.ts

@@ -10,7 +10,7 @@ import { invariant } from '@dxos/invariant';
 import { log } from '@dxos/log';
 import { trace } from '@dxos/tracing';
 
-import { type WavConfig, type AudioChunk, type AudioRecorder } from './audio-recorder';
+import { type AudioChunk, type AudioRecorder, type WavConfig } from './audio-recorder';
 
 let initializingPromise: Promise<void> | undefined;
 
@@ -19,7 +19,7 @@ const initializeExtendableMediaRecorder = async () => {
   await register(await connect());
 };
 
-export type MediaStreamRecorderParams = {
+export type MediaStreamRecorderProps = {
   mediaStreamTrack: MediaStreamTrack;
   config: {
     interval: number;
@@ -37,12 +37,12 @@ export class MediaStreamRecorder implements AudioRecorder {
   /**
    * Default MediaRecorder implementation do not support wav encoding.
    */
-  private readonly _config: MediaStreamRecorderParams['config'];
+  private readonly _config: MediaStreamRecorderProps['config'];
   private _mediaRecorder?: IMediaRecorder = undefined;
   private _header?: Uint8Array = undefined;
   private _onChunk?: (chunk: AudioChunk) => void;
 
-  constructor({ mediaStreamTrack, config }: MediaStreamRecorderParams) {
+  constructor({ mediaStreamTrack, config }: MediaStreamRecorderProps) {
     this._mediaStreamTrack = mediaStreamTrack;
     this._config = config;
   }

package/src/transcriber/transcriber.browser.test.ts

@@ -4,14 +4,15 @@
 
 import fs from 'node:fs';
 import path from 'node:path';
+
 import { describe, expect, test } from 'vitest';
 import { WaveFile } from 'wavefile';
 
 import { Trigger } from '@dxos/async';
 import { log } from '@dxos/log';
-import { type DataType } from '@dxos/schema';
 import { openAndClose } from '@dxos/test-utils';
-import { trace, TRACE_PROCESSOR } from '@dxos/tracing';
+import { TRACE_PROCESSOR, trace } from '@dxos/tracing';
+import { type ContentBlock } from '@dxos/types';
 
 import { type AudioChunk, type AudioRecorder, Transcriber } from '../transcriber';
 import { mergeFloat64Arrays } from '../util';
@@ -93,7 +94,7 @@ describe.skip('Transcriber', () => {
   });
 
   test.skip('transcription of audio recording', { timeout: 10_000 }, async () => {
-    const trigger = new Trigger<DataType.MessageBlock.Transcription[]>({ autoReset: true });
+    const trigger = new Trigger<ContentBlock.Transcript[]>({ autoReset: true });
     const recorder = new MockAudioRecorder({
       buffer: await readFile('test.wav'),
       chunkDuration: 3_000,
@@ -127,7 +128,7 @@ describe.skip('Transcriber', () => {
   });
 
   test.skip('transcription of audio recording with overlapping chunks', { timeout: 20_000 }, async () => {
-    const trigger = new Trigger<DataType.MessageBlock.Transcription[]>({ autoReset: true });
+    const trigger = new Trigger<ContentBlock.Transcript[]>({ autoReset: true });
     const recorder = new MockAudioRecorder({
       buffer: await readFile('test.wav'),
       chunkDuration: 3_000,
@@ -212,7 +213,7 @@ class MockAudioRecorder implements AudioRecorder {
 
   @trace.span()
   async start(): Promise<void> {
-    log.info('start mock audio recorder', { wavParams: this.wav.fmt });
+    log.info('start mock audio recorder', { wavProps: this.wav.fmt });
     const now = Date.now();
     const samples = this.wav.getSamples();
     const sampleRate = this.wavConfig.sampleRate;

package/src/transcriber/transcriber.ts

@@ -7,13 +7,14 @@ import { WaveFile } from 'wavefile';
 import { DeferredTask, Trigger } from '@dxos/async';
 import { type Context, LifecycleState, Resource } from '@dxos/context';
 import { log } from '@dxos/log';
-import { type DataType } from '@dxos/schema';
 import { trace } from '@dxos/tracing';
+import { type ContentBlock } from '@dxos/types';
 
-import { type AudioRecorder, type AudioChunk } from './audio-recorder';
 import { TRANSCRIPTION_URL } from '../types';
 import { mergeFloat64Arrays } from '../util';
 
+import { type AudioChunk, type AudioRecorder } from './audio-recorder';
+
 type WhisperWord = {
   word: string;
 
@@ -62,14 +63,14 @@ export type TranscribeConfig = {
   prefixBufferChunksAmount: number;
 };
 
-export type TranscriberParams = {
+export type TranscriberProps = {
   config: TranscribeConfig;
   recorder: AudioRecorder;
   /**
    * Callback to handle the transcribed segments, after all segment transformers are applied.
    * @param segments - The transcribed segments.
    */
-  onSegments: (segments: DataType.MessageBlock.Transcription[]) => Promise<void>;
+  onSegments: (segments: ContentBlock.Transcript[]) => Promise<void>;
 };
 
 /**
@@ -85,12 +86,12 @@ export class Transcriber extends Resource {
 
   private readonly _config: TranscribeConfig;
   private readonly _recorder: AudioRecorder;
-  private readonly _onSegments: TranscriberParams['onSegments'];
+  private readonly _onSegments: TranscriberProps['onSegments'];
 
   private _recording = false;
   private _transcribeTask?: DeferredTask = undefined;
 
-  constructor({ config, recorder, onSegments }: TranscriberParams) {
+  constructor({ config, recorder, onSegments }: TranscriberProps) {
     super();
     this._config = config;
     this._recorder = recorder;
@@ -220,10 +221,7 @@ export class Transcriber extends Resource {
     return segments;
   }
 
-  private _alignSegments(
-    segments: WhisperSegment[],
-    originalChunks: AudioChunk[],
-  ): DataType.MessageBlock.Transcription[] {
+  private _alignSegments(segments: WhisperSegment[], originalChunks: AudioChunk[]): ContentBlock.Transcript[] {
     // Absolute zero for all relative timestamps in the segments.
     const zeroTimestamp = originalChunks.at(0)!.timestamp;
 
@@ -252,7 +250,7 @@ export class Transcriber extends Resource {
 
     // Add absolute timestamp to each segment.
     return filteredSegments.map((segment) => ({
-      type: 'transcription',
+      _tag: 'transcript',
       started: new Date(zeroTimestamp + segment.start * 1_000).toISOString(),
       text: segment.text.trim(),
     }));