Variable manifestConst

Type Declaration

• Readonlyid: "transcription"

• Readonlyrequires: readonly ["modelHost", "catalog"]

• Readonlyroutes: readonly [
      {
          id: "transcriptions";
          method: "POST";
          mount: "root";
          path: "/v1/audio/transcriptions";
          response: "json";
          summary: "Transcribe an uploaded clip (multipart or JSON body); text, json, verbose_json, srt or vtt, or an SSE stream";
      },
      {
          id: "translations";
          method: "POST";
          mount: "root";
          path: "/v1/audio/translations";
          response: "json";
          summary: "Translate an uploaded clip to English";
      },
      {
          id: "unload";
          method: "POST";
          mount: "root";
          path: "/admin/transcription/unload";
          response: "json";
          summary: "Page the Whisper weights out now; reports the residency counters";
      },
      {
          id: "session-create";
          method: "POST";
          mount: "root";
          path: "/v1/audio/sessions";
          response: "json";
          summary: "Open a streaming dictation session";
      },
      {
          id: "session-audio";
          method: "POST";
          mount: "root";
          path: "/v1/audio/sessions/:id/audio";
          response: "json";
          summary: "Append a chunk of 16 kHz float32 PCM (or an encoded clip) to a session";
      },
      {
          id: "session-finish";
          method: "POST";
          mount: "root";
          path: "/v1/audio/sessions/:id/finish";
          response: "json";
          summary: "Transcribe the remainder and close the session";
      },
      {
          id: "session-delete";
          method: "DELETE";
          mount: "root";
          path: "/v1/audio/sessions/:id";
          response: "json";
          summary: "Close a session that was abandoned";
      },
  ]

• Readonlysummary: "Speech-to-text with a local Whisper model: OpenAI-compatible audio routes, streaming dictation sessions, and the transcribe and dictate verbs."

• Readonlytitle: "Transcription"

• Readonlyverbs: readonly [
      {
          name: "transcribe";
          options: readonly [
              {
                  name: "model";
                  summary: "Whisper model directory or cached query (overrides the positional query) [default: the first downloaded whisper checkpoint]";
                  type: "string";
              },
              {
                  name: "query";
                  summary: "Cached Whisper query when no positional/model override is supplied";
                  type: "string";
              },
              {
                  name: "language";
                  summary: "ISO code or name (en, japanese); auto or omitted detects from the first 30 s";
                  type: "string";
              },
              {
                  name: "task";
                  summary: "transcribe | translate (to English) [default: transcribe]";
                  type: "string";
              },
              {
                  name: "beam-size";
                  summary: "Beam search width [default: greedy]";
                  type: "number";
              },
              {
                  name: "temperature";
                  summary: "One sampling temperature; omitted runs the (0, 0.2, …, 1.0) fallback ladder like mlx-whisper";
                  type: "number";
              },
              {
                  name: "no-fallback";
                  summary: "Temperature 0 only, no retry ladder";
                  type: "boolean";
              },
              {
                  name: "prompt";
                  summary: "Initial prompt (vocabulary hints / style)";
                  type: "string";
              },
              {
                  name: "no-timestamps";
                  summary: "Decode text only (<|notimestamps|>)";
                  type: "boolean";
              },
              {
                  name: "no-condition";
                  summary: "Do not condition each window on the previous text";
                  type: "boolean";
              },
              {
                  name: "vad";
                  summary: "Silero VAD gate: no detected speech prints an empty result and Whisper never runs (weights: ggml-org/whisper-vad in the HF cache)";
                  type: "boolean";
              },
              {
                  name: "vad-threshold";
                  summary: "Speech probability threshold for --vad [default: 0.5]";
                  type: "number";
              },
              {
                  name: "vad-model";
                  summary: "Explicit Silero ggml path for --vad";
                  type: "string";
              },
              {
                  name: "vad-trim";
                  summary: "Accepted with --vad; the clip is still transcribed whole, as in main";
                  type: "boolean";
              },
              {
                  name: "word-timestamps";
                  summary: "Word-level timestamps (verbose_json words)";
                  type: "boolean";
              },
              {
                  name: "faithful";
                  summary: "Run the oracle-parity graph instead of the fast path";
                  type: "boolean";
              },
              {
                  name: "audio-ctx";
                  summary: "Lab: encode only n of the 1500 encoder positions (whisper.cpp -ac); degrades below ~1024";
                  type: "number";
              },
              {
                  name: "format";
                  summary: "text | json | verbose_json | srt | vtt [default: text]";
                  type: "string";
              },
              {
                  name: "verbose";
                  summary: "Print each segment with its time range as it decodes, then a realtime-factor summary (stderr)";
                  type: "boolean";
              },
          ];
          positional: readonly [
              {
                  name: "audio-file";
                  required: true;
                  summary: "The clip to transcribe";
              },
              {
                  name: "query";
                  summary: "The Whisper checkpoint (a directory or a cached query)";
              },
          ];
          summary: "Speech-to-text from an audio file with a local Whisper model (no server)";
          usage: "usage: mlx-bun transcribe <audio-file> [query] [--language en] [--format text|json|verbose_json|srt|vtt]";
      },
      {
          name: "dictate";
          options: readonly [
              {
                  name: "model";
                  summary: "Whisper model directory or cached query for in-process transcription (overrides the positional query) [default: the first downloaded whisper checkpoint]";
                  type: "string";
              },
              {
                  name: "query";
                  summary: "Cached Whisper query when no positional/model override is supplied";
                  type: "string";
              },
              {
                  name: "server";
                  summary: "Use a running mlx-bun server's /v1/audio/sessions instead of loading the model here";
                  type: "string";
              },
              {
                  default: 61;
                  name: "hotkey";
                  optionalValue: true;
                  summary: "Hold a key to talk instead of Enter toggling: macOS virtual keycode, value optional [default: 61 = Right Option]; needs Input Monitoring for your terminal";
                  type: "number";
              },
              {
                  name: "language";
                  summary: "Language code; auto detects [default: en]";
                  type: "string";
              },
              {
                  name: "beam-size";
                  summary: "Beam search width [default: greedy]";
                  type: "number";
              },
              { name: "prompt"; summary: "Initial prompt"; type: "string" },
              {
                  name: "vocabulary";
                  summary: "Comma-separated hint terms, fitted into the prompt budget";
                  type: "string";
              },
              {
                  name: "no-vad";
                  summary: "Skip the Silero gate (by default silence never runs Whisper)";
                  type: "boolean";
              },
              {
                  name: "idle-unload";
                  summary: "Release the weights after this many idle seconds; 0 = right after each take [default: 30]";
                  type: "number";
              },
              { name: "resident"; summary: "Keep the weights loaded"; type: "boolean" },
              {
                  name: "copy";
                  summary: "Copy the transcript to the clipboard (pbcopy)";
                  type: "boolean";
              },
              {
                  name: "type";
                  summary: "Type it into the frontmost app via System Events keystrokes (needs Accessibility for your terminal)";
                  type: "boolean";
              },
              {
                  name: "type-delay";
                  summary: "Seconds to wait before typing so you can Cmd-Tab [default: 1 in Enter mode, 0 with --hotkey]";
                  type: "number";
              },
          ];
          positional: readonly [
              {
                  name: "query";
                  summary: "The Whisper checkpoint (a directory or a cached query)";
              },
          ];
          summary: "Push-to-talk: stream your microphone into Whisper, print/copy/type the text";
      },
  ]