Workspace / Studio Session
Local Inference

Live Transcription

STREAMING
00:04:21 .8

04:12 So when we consider the architecture of the new model, we need to ensure that the quantization doesn't heavily impact the perplexity score...

04:18 And that's exactly why we are implementing the...

Audio Interface

Format
48,000 Hz / 24-bit

Input Levels

-60 -30 -12 0 dB
-42 dB
Auto Gain Control

Inference Engine

Whisper Large-v3 Q4_K_M
2.9 GB • Requires 4.1 GB RAM
Active
Distil-Whisper English
Downloading... 45% (850 MB / 1.8 GB)
Whisper Tiny.en
150 MB • Ultra-fast • Lower accuracy

Pipeline & Processing

Offline Mode
Speaker Diarization
Auto Punctuation
Trim Silence
Mistral-7B

System Output

14:32:01.045 [INFO] Audio input initialized: Universal Audio Volt (48000Hz, 1ch)
14:32:01.112 [INFO] Loading model 'whisper-large-v3-q4_k_m.bin' to Metal GPU...
14:32:02.850 [INFO] Model loaded successfully in 1.73s. VRAM usage: 4120 MB.
14:32:03.001 [INFO] Inference engine ready. Awaiting audio stream...
14:32:04.220 [STREAM] Started recording session ID: sess_94x8a2
14:32:05.105 [PIPELINE] Chunk 1 processed (RTF: 0.12x)
14:32:06.310 [PIPELINE] Chunk 2 processed (RTF: 0.11x)