first commit

2026-06-04 18:10:52 +07:00
commit b5c083e06f
105 changed files with 8172 additions and 0 deletions
@@ -0,0 +1,55 @@
+api:
+  addr: "0.0.0.0:6183"
+  models_dir: "./models"
+  cache_dir: "./cache"
+  # Open WebUI / OpenAI STT: model id when client sends whisper-1 (e.g. ggml-large-v3-turbo)
+  default_model: ggml-large-v3-turbo
+  threads: 16
+  language: ru
+  transcript:
+    pause_gap_sec: 1.5
+    speaker_label: "Спикер"
+  default_speakers: false
+  debug: false
+  speedup: false
+  translate: false
+  prompt: ""
+  max_context: 32
+  beam_size: 5
+  entropy_thold: 2.4
+  vad:
+    enabled: false
+    model: vad/vad.bin
+    threshold: 0.5
+    min_speech_duration_ms: 250
+    min_silence_duration_ms: 100
+    speech_pad_ms: 30
+    samples_overlap: 0.1
+  default_punctuation: true
+  default_async: true
+  garbage:
+    - "*выбая*"
+
+# transcode: pure Go (wav, mp3, flac, ogg, m4a, mp4, aac) — no ffmpeg required
+transcode: {}
+
+diarization:
+  enabled: false
+  model_dir: ./models/diarization
+  segmentation_model: pyannote-segmentation-3-0/model.onnx
+  embedding_model: 3dspeaker_speech_eres2net_base_sv_zh-cn_3dspeaker_16k.onnx
+  num_threads: 2
+  num_clusters: 0
+  clustering_threshold: 0.5
+
+punctuation:
+  enabled: true
+  default_on: true
+  # off | heuristic | xlm | sherpa | sherpa-online | http
+  engine: xlm
+  model_dir: ./models/punctuation/xlm-roberta
+  model_file: model.onnx
+  sp_model: sp.model
+  config_file: config.yaml
+  apply_sbd: true
+  num_threads: 2