whisper-stream doesn't do diarization, while whisper-cli does

[dimon777]

Hi!

I am testing diarization feature and it is not working in whisper-stream at all. Here is simple test using this podcast: (sample file is here)

whisper-cli (works):

./build/bin/whisper-cli -f ./output.mp3 -m ./models/ggml-small.en-tdrz.bin -tdrz -pc

wisper-stream (doesn't work):

# terminal 1: 
ffplay ./output.mp3

# terminal 2:
./build/bin/whisper-stream -m ./models/ggml-small.en-tdrz.bin -t 8 -tdrz -c 1 -kc

[nilokmiss-max]

Hi dimon777, Thank you for reporting this issue regarding the diarization feature in whisper-stream. I am looking into why the diarization is working in whisper-cli but failing to function in whisper-stream as shown in your test with the podcast. Could you please provide the specific commands or parameters you used for both whisper-cli and whisper-stream? This will help in reproducing the issue more accurately. Best regards, nilok Miss

…

On Tue, Mar 17, 2026, 9:00 PM dimon777 ***@***.***> wrote: *dimon777* created an issue (ggml-org/whisper.cpp#3712) <#3712> Hi! I am testing diarization feature and it is not working in whisper-stream at all. Here is simple test using this podcast <https://www.youtube.com/watch?v=azvA2Bn2aXw>: whisper-cli (works): image.png (view on web) <https://github.com/user-attachments/assets/4019cf54-8376-4d20-922b-473ad654ff97> wisper-stream (doesn't work): image.png (view on web) <https://github.com/user-attachments/assets/b5f2bb0d-c2a3-4e73-9a75-3be7af93688e> — Reply to this email directly, view it on GitHub <#3712>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/B7VEKIQE3ZNXZRVFLZGCP5T4RH7LFAVCNFSM6AAAAACWV2JECSVHI2DSMVQWIX3LMV43ASLTON2WKOZUGA4TEMRRHEZDEMI> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[dimon777]

Hi @nilokmiss-max

All commands are there in the original post. The "-c 1" in whisper-stream points to specific audio channel on my Mac laptop to capture ffplay output. This is Mult-Output Device configured via:

$ brew install blackhole-16ch

Then:

• Open Audio MIDI Setup (found in Applications > Utilities).
• Click the + at the bottom left and create a Multi-Output Device.
• Check both Built-in Output (so you can still hear the sound) and BlackHole 16ch.
• Right-click the Multi-Output Device and select "Use This Device For Sound Output".

And identified with:

ffmpeg -f avfoundation -list_devices true -i ""
[AVFoundation indev @ 0xb77020000] AVFoundation video devices:
[AVFoundation indev @ 0xb77020000] [0] onn. USB 2.0 webcam
[AVFoundation indev @ 0xb77020000] [1] FaceTime HD Camera
[AVFoundation indev @ 0xb77020000] [2] Capture screen 0
[AVFoundation indev @ 0xb77020000] AVFoundation audio devices:
[AVFoundation indev @ 0xb77020000] [0] onn. USB 2.0 webcam audio
[AVFoundation indev @ 0xb77020000] [1] BlackHole 16ch                                <========= :1 device to stream from
[AVFoundation indev @ 0xb77020000] [2] MacBook Pro Microphone