Incorrect log output when --dtw is enabled

[rotemdan]

Without --dtw:

whisper-cli test2.wav

whisper_init_from_file_with_params_no_state: loading model from 'models/ggml-base.en.bin'
whisper_init_with_params_no_state: use gpu    = 1
whisper_init_with_params_no_state: flash attn = 0
whisper_init_with_params_no_state: gpu_device = 0
whisper_init_with_params_no_state: dtw        = 0
whisper_init_with_params_no_state: devices    = 1
whisper_init_with_params_no_state: backends   = 1
whisper_model_load: loading model
whisper_model_load: n_vocab       = 51864
whisper_model_load: n_audio_ctx   = 1500
whisper_model_load: n_audio_state = 512
whisper_model_load: n_audio_head  = 8
whisper_model_load: n_audio_layer = 6
whisper_model_load: n_text_ctx    = 448
whisper_model_load: n_text_state  = 512
whisper_model_load: n_text_head   = 8
whisper_model_load: n_text_layer  = 6
whisper_model_load: n_mels        = 80
whisper_model_load: ftype         = 1
whisper_model_load: qntvr         = 0
whisper_model_load: type          = 2 (base)
whisper_model_load: adding 1607 extra tokens
whisper_model_load: n_langs       = 99
whisper_model_load:          CPU total size =   147.37 MB
whisper_model_load: model size    =  147.37 MB
whisper_backend_init_gpu: no GPU found
whisper_init_state: kv self size  =    6.29 MB
whisper_init_state: kv cross size =   18.87 MB
whisper_init_state: kv pad  size  =    3.15 MB
whisper_init_state: compute buffer (conv)   =   16.26 MB
whisper_init_state: compute buffer (encode) =   85.86 MB
whisper_init_state: compute buffer (cross)  =    4.65 MB
whisper_init_state: compute buffer (decode) =   96.35 MB

system_info: n_threads = 4 / 8 | WHISPER : COREML = 0 | OPENVINO = 0 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | OPENMP = 1 | AARCH64_REPACK = 1 |

main: processing 'test2.wav' (5043374 samples, 315.2 sec), 4 threads, 1 processors, 5 beams + best of 5, lang = en, task = transcribe, timestamps = 1 ...


[00:00:00.000 --> 00:00:02.520]   (upbeat music)
[00:00:02.520 --> 00:00:03.360]   Guess what?
[00:00:03.360 --> 00:00:09.440]   So, remember how you created a fire pool for me.
[00:00:09.440 --> 00:00:11.560]   - Kind of remember that.
[00:00:11.560 --> 00:00:12.720]   - Yeah, from the...
[00:00:12.720 --> 00:00:13.560]   - Wasn't it real?
[00:00:13.560 --> 00:00:14.840]   - Up on the top of the mezzanine down.
[00:00:14.840 --> 00:00:16.720]   No, no, you made a fire pool.
[00:00:16.720 --> 00:00:17.720]   - Is it real?
[00:00:17.720 --> 00:00:19.200]   Is it an illusion?
[00:00:19.200 --> 00:00:21.400]   Only the editors will know.
[00:00:21.400 --> 00:00:22.600]   - By the end of the day,
[00:00:22.600 --> 00:00:25.360]   I want you to have a real fire pool installed
[00:00:25.360 --> 00:00:26.640]   that I can slide down.
[00:00:26.640 --> 00:00:28.640]   - You know, I have things to do, right?
[00:00:28.640 --> 00:00:30.960]   - Yeah, they look really, really important.
[00:00:30.960 --> 00:00:33.560]   All right, cool, I'm glad we agree.
[00:00:33.560 --> 00:00:38.220]   (upbeat music)
[00:00:38.220 --> 00:00:44.080]   Private internet access is the VPN service
[00:00:44.080 --> 00:00:46.180]   that encrypts all of your internet traffic
[00:00:46.180 --> 00:00:48.640]   and uses a safe, protected IP.
[00:00:48.640 --> 00:00:51.160]   And it's got tons of other useful features as well.
[00:00:51.160 --> 00:00:53.920]   Check it out now at the link below.
[00:00:53.920 --> 00:00:56.120]   All right, so, plan A.
[00:00:56.120 --> 00:00:58.600]   How long have you spent developing this plan?
[00:00:58.600 --> 00:01:00.960]   - Well, extend this, we'll have this thing here.
[00:01:00.960 --> 00:01:03.240]   We'll have people holding onto this.
[00:01:03.240 --> 00:01:04.080]   - Which people?
[00:01:04.080 --> 00:01:05.600]   - We can find people on the street
[00:01:05.600 --> 00:01:10.400]   and we'll also clamp it onto the railing up there.
[00:01:10.400 --> 00:01:13.360]   - Can I reveal something that you probably don't know?
[00:01:13.360 --> 00:01:15.320]   - There's pieces of information you don't have.
[00:01:15.320 --> 00:01:18.400]   - Okay, so when we put in these stairs,
[00:01:18.400 --> 00:01:22.760]   we'd ever actually intended for anyone to walk down them.
[00:01:23.920 --> 00:01:25.420]   These are the way up.
[00:01:25.420 --> 00:01:31.440]   Whoa, I knew that.
[00:01:31.440 --> 00:01:34.880]   Firepool installation, baby.
[00:01:34.880 --> 00:01:40.600]   Wow, this is a little nerve-wracking, mind me.
[00:01:40.600 --> 00:01:44.360]   So, instead of having this long pole go this way,
[00:01:44.360 --> 00:01:47.160]   like this way, we can attach it
[00:01:47.160 --> 00:01:50.880]   to several of these posts instead.
[00:01:50.880 --> 00:01:53.040]   Okay, so let's saw the Yolo plan.
[00:01:54.000 --> 00:01:55.400]   What do you mean the Yolo plan?
[00:01:55.400 --> 00:01:58.120]   - I want something simpler and faster than this.
[00:01:58.120 --> 00:01:59.360]   - Are you serious?
[00:01:59.360 --> 00:02:02.960]   - Maybe fewer people.
[00:02:02.960 --> 00:02:06.520]   One person will attach weights to it.
[00:02:06.520 --> 00:02:08.760]   We'll tape weights to this and rope.
[00:02:08.760 --> 00:02:09.600]   We have rope.
[00:02:09.600 --> 00:02:10.600]   - All right.
[00:02:10.600 --> 00:02:11.520]   - We're not doing this.
[00:02:11.520 --> 00:02:12.360]   - I'm not doing this.
[00:02:12.360 --> 00:02:13.880]   - Approached. - No, no.
[00:02:13.880 --> 00:02:15.800]   - You're not serious about the Yolo, right?
[00:02:15.800 --> 00:02:17.000]   - Yeah, yeah, Yolo plan.
[00:02:17.000 --> 00:02:18.680]   - No, you're not, no, no, you're not.
[00:02:18.680 --> 00:02:19.520]   - Let's give him stuff out of the way.
[00:02:19.520 --> 00:02:21.560]   - We're not Yoloing, I'm gonna tell the bond.
[00:02:21.560 --> 00:02:24.160]   - I don't wanna lose my job because he died.
[00:02:24.160 --> 00:02:27.760]   - The fire pole is going in today.
[00:02:27.760 --> 00:02:29.840]   - What happened to the safety concerns?
[00:02:29.840 --> 00:02:31.000]   Do we have insurance for this?
[00:02:31.000 --> 00:02:33.600]   - I think if anyone came in and inspected our implementation,
[00:02:33.600 --> 00:02:35.640]   they'd be really happy with it.
[00:02:35.640 --> 00:02:37.600]   - Oh yeah? - Yeah.
[00:02:37.600 --> 00:02:39.280]   - We'll have a cross brace.
[00:02:39.280 --> 00:02:40.680]   We'll attach it to the post.
[00:02:40.680 --> 00:02:42.840]   - Yeah, this all sounds incredibly stupid.
[00:02:42.840 --> 00:02:46.360]   - I feel like I should be wearing a hard hat.
[00:02:46.360 --> 00:02:50.200]   - Yeah, do we have a hard hat?
[00:02:51.200 --> 00:02:52.600]   (beep)

[... truncated ...]

With --dtw:

whisper-cli test2.wav --dtw base.en

whisper_init_from_file_with_params_no_state: loading model from 'models/ggml-base.en.bin'
whisper_init_with_params_no_state: use gpu    = 1
whisper_init_with_params_no_state: flash attn = 0
whisper_init_with_params_no_state: gpu_device = 0
whisper_init_with_params_no_state: dtw        = 1
whisper_init_with_params_no_state: devices    = 1
whisper_init_with_params_no_state: backends   = 1
whisper_model_load: loading model
whisper_model_load: n_vocab       = 51864
whisper_model_load: n_audio_ctx   = 1500
whisper_model_load: n_audio_state = 512
whisper_model_load: n_audio_head  = 8
whisper_model_load: n_audio_layer = 6
whisper_model_load: n_text_ctx    = 448
whisper_model_load: n_text_state  = 512
whisper_model_load: n_text_head   = 8
whisper_model_load: n_text_layer  = 6
whisper_model_load: n_mels        = 80
whisper_model_load: ftype         = 1
whisper_model_load: qntvr         = 0
whisper_model_load: type          = 2 (base)
whisper_model_load: adding 1607 extra tokens
whisper_model_load: n_langs       = 99
whisper_model_load:          CPU total size =   147.37 MB
whisper_model_load: model size    =  147.37 MB
whisper_backend_init_gpu: no GPU found
whisper_init_state: kv self size  =    6.29 MB
whisper_init_state: kv cross size =   18.87 MB
whisper_init_state: kv pad  size  =    3.15 MB
whisper_init_state: alignment heads masks size = 160 B
whisper_init_state: compute buffer (conv)   =   16.26 MB
whisper_init_state: compute buffer (encode) =   85.86 MB
whisper_init_state: compute buffer (cross)  =    4.65 MB
whisper_init_state: compute buffer (decode) =  109.79 MB

system_info: n_threads = 4 / 8 | WHISPER : COREML = 0 | OPENVINO = 0 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | OPENMP = 1 | AARCH64_REPACK = 1 |

main: processing 'test2.wav' (5043374 samples, 315.2 sec), 4 threads, 1 processors, 5 beams + best of 5, lang = en, task = transcribe, timestamps = 1 ...

[00:00:26.640 --> 00:00:28.640]   - You know, I have things to do, right?
[00:00:25.360 --> 00:00:26.640]   that I can slide down.
[00:00:26.640 --> 00:00:28.640]   - You know, I have things to do, right?
[00:00:22.600 --> 00:00:25.360]   I want you to have a real fire pool installed
[00:00:25.360 --> 00:00:26.640]   that I can slide down.
[00:00:26.640 --> 00:00:28.640]   - You know, I have things to do, right?
[00:00:21.400 --> 00:00:22.600]   - By the end of the day,
[00:00:22.600 --> 00:00:25.360]   I want you to have a real fire pool installed
[00:00:25.360 --> 00:00:26.640]   that I can slide down.
[00:00:26.640 --> 00:00:28.640]   - You know, I have things to do, right?
[00:00:19.200 --> 00:00:21.400]   Only the editors will know.
[00:00:21.400 --> 00:00:22.600]   - By the end of the day,
[00:00:22.600 --> 00:00:25.360]   I want you to have a real fire pool installed
[00:00:25.360 --> 00:00:26.640]   that I can slide down.
[00:00:26.640 --> 00:00:28.640]   - You know, I have things to do, right?
[00:00:17.720 --> 00:00:19.200]   Is it an illusion?
[00:00:19.200 --> 00:00:21.400]   Only the editors will know.
[00:00:21.400 --> 00:00:22.600]   - By the end of the day,
[00:00:22.600 --> 00:00:25.360]   I want you to have a real fire pool installed
[00:00:25.360 --> 00:00:26.640]   that I can slide down.
[00:00:26.640 --> 00:00:28.640]   - You know, I have things to do, right?
[00:00:16.720 --> 00:00:17.720]   - Is it real?
[00:00:17.720 --> 00:00:19.200]   Is it an illusion?
[00:00:19.200 --> 00:00:21.400]   Only the editors will know.
[00:00:21.400 --> 00:00:22.600]   - By the end of the day,
[00:00:22.600 --> 00:00:25.360]   I want you to have a real fire pool installed
[00:00:25.360 --> 00:00:26.640]   that I can slide down.
[00:00:26.640 --> 00:00:28.640]   - You know, I have things to do, right?
[00:00:14.840 --> 00:00:16.720]   No, no, you made a fire pool.
[00:00:16.720 --> 00:00:17.720]   - Is it real?
[00:00:17.720 --> 00:00:19.200]   Is it an illusion?
[00:00:19.200 --> 00:00:21.400]   Only the editors will know.
[00:00:21.400 --> 00:00:22.600]   - By the end of the day,
[00:00:22.600 --> 00:00:25.360]   I want you to have a real fire pool installed
[00:00:25.360 --> 00:00:26.640]   that I can slide down.
[00:00:26.640 --> 00:00:28.640]   - You know, I have things to do, right?
[00:00:13.560 --> 00:00:14.840]   - Up on the top of the mezzanine down.
[00:00:14.840 --> 00:00:16.720]   No, no, you made a fire pool.
[00:00:16.720 --> 00:00:17.720]   - Is it real?
[00:00:17.720 --> 00:00:19.200]   Is it an illusion?
[00:00:19.200 --> 00:00:21.400]   Only the editors will know.
[00:00:21.400 --> 00:00:22.600]   - By the end of the day,
[00:00:22.600 --> 00:00:25.360]   I want you to have a real fire pool installed
[00:00:25.360 --> 00:00:26.640]   that I can slide down.
[00:00:26.640 --> 00:00:28.640]   - You know, I have things to do, right?
[00:00:12.720 --> 00:00:13.560]   - Wasn't it real?
[00:00:13.560 --> 00:00:14.840]   - Up on the top of the mezzanine down.
[00:00:14.840 --> 00:00:16.720]   No, no, you made a fire pool.
[00:00:16.720 --> 00:00:17.720]   - Is it real?
[00:00:17.720 --> 00:00:19.200]   Is it an illusion?
[00:00:19.200 --> 00:00:21.400]   Only the editors will know.
[00:00:21.400 --> 00:00:22.600]   - By the end of the day,
[00:00:22.600 --> 00:00:25.360]   I want you to have a real fire pool installed
[00:00:25.360 --> 00:00:26.640]   that I can slide down.
[00:00:26.640 --> 00:00:28.640]   - You know, I have things to do, right?
[00:00:11.560 --> 00:00:12.720]   - Yeah, from the...
[00:00:12.720 --> 00:00:13.560]   - Wasn't it real?
[00:00:13.560 --> 00:00:14.840]   - Up on the top of the mezzanine down.
[00:00:14.840 --> 00:00:16.720]   No, no, you made a fire pool.
[00:00:16.720 --> 00:00:17.720]   - Is it real?
[00:00:17.720 --> 00:00:19.200]   Is it an illusion?
[00:00:19.200 --> 00:00:21.400]   Only the editors will know.
[00:00:21.400 --> 00:00:22.600]   - By the end of the day,
[00:00:22.600 --> 00:00:25.360]   I want you to have a real fire pool installed
[00:00:25.360 --> 00:00:26.640]   that I can slide down.
[00:00:26.640 --> 00:00:28.640]   - You know, I have things to do, right?
[00:00:09.440 --> 00:00:11.560]   - Kind of remember that.
[00:00:11.560 --> 00:00:12.720]   - Yeah, from the...
[00:00:12.720 --> 00:00:13.560]   - Wasn't it real?
[00:00:13.560 --> 00:00:14.840]   - Up on the top of the mezzanine down.
[00:00:14.840 --> 00:00:16.720]   No, no, you made a fire pool.
[00:00:16.720 --> 00:00:17.720]   - Is it real?
[00:00:17.720 --> 00:00:19.200]   Is it an illusion?
[00:00:19.200 --> 00:00:21.400]   Only the editors will know.
[00:00:21.400 --> 00:00:22.600]   - By the end of the day,
[00:00:22.600 --> 00:00:25.360]   I want you to have a real fire pool installed
[00:00:25.360 --> 00:00:26.640]   that I can slide down.
[00:00:26.640 --> 00:00:28.640]   - You know, I have things to do, right?
[00:00:03.360 --> 00:00:09.440]   So, remember how you created a fire pool for me.
[00:00:09.440 --> 00:00:11.560]   - Kind of remember that.
[00:00:11.560 --> 00:00:12.720]   - Yeah, from the...
[00:00:12.720 --> 00:00:13.560]   - Wasn't it real?
[00:00:13.560 --> 00:00:14.840]   - Up on the top of the mezzanine down.
[00:00:14.840 --> 00:00:16.720]   No, no, you made a fire pool.
[00:00:16.720 --> 00:00:17.720]   - Is it real?
[00:00:17.720 --> 00:00:19.200]   Is it an illusion?
[00:00:19.200 --> 00:00:21.400]   Only the editors will know.
[00:00:21.400 --> 00:00:22.600]   - By the end of the day,
[00:00:22.600 --> 00:00:25.360]   I want you to have a real fire pool installed
[00:00:25.360 --> 00:00:26.640]   that I can slide down.
[00:00:26.640 --> 00:00:28.640]   - You know, I have things to do, right?
[00:00:02.520 --> 00:00:03.360]   Guess what?
[00:00:03.360 --> 00:00:09.440]   So, remember how you created a fire pool for me.
[00:00:09.440 --> 00:00:11.560]   - Kind of remember that.
[00:00:11.560 --> 00:00:12.720]   - Yeah, from the...
[00:00:12.720 --> 00:00:13.560]   - Wasn't it real?
[00:00:13.560 --> 00:00:14.840]   - Up on the top of the mezzanine down.
[00:00:14.840 --> 00:00:16.720]   No, no, you made a fire pool.
[00:00:16.720 --> 00:00:17.720]   - Is it real?
[00:00:17.720 --> 00:00:19.200]   Is it an illusion?
[00:00:19.200 --> 00:00:21.400]   Only the editors will know.
[00:00:21.400 --> 00:00:22.600]   - By the end of the day,
[00:00:22.600 --> 00:00:25.360]   I want you to have a real fire pool installed
[00:00:25.360 --> 00:00:26.640]   that I can slide down.
[00:00:26.640 --> 00:00:28.640]   - You know, I have things to do, right?


whisper_print_timings:     load time =   183.36 ms
whisper_print_timings:     fallbacks =   2 p /   9 h
whisper_print_timings:      mel time =   192.15 ms
whisper_print_timings:   sample time =  5801.55 ms /  9386 runs (    0.62 ms per run)
whisper_print_timings:   encode time = 13922.25 ms /    12 runs ( 1160.19 ms per run)
whisper_print_timings:   decode time =  1053.59 ms /   197 runs (    5.35 ms per run)
whisper_print_timings:   batchd time = 18152.11 ms /  9123 runs (    1.99 ms per run)
whisper_print_timings:   prompt time =  4975.55 ms /  3767 runs (    1.32 ms per run)
whisper_print_timings:    total time = 44862.12 ms

The issue did not happen on older versions (one I had was from October 2024).

Seems to be a regression on some sort. It happens both on latest code and on latest stable build (1.7.5).

The t_dtw timestamps themselves and the standard timestamps in the JSON object do seem to be accurate (t_dtw timestamps are converted to timestamps in Echogarden when this option is enabled, so I can test them directly). The problem may be more about the log output.