[BUG] Recording video with audio=True results in "Non-monotonous DTS in output stream"

[guighub]

Describe the bug Trying to record video with audio from a USB microphone results in a lot of "Non-monotonous DTS in output stream" errors from ffmpeg and preventing my script from running, I also tried using FfmpegOutput but got the same result.

To Reproduce

config = cam.create_video_configuration(main={"size": video_resolution}, lores={"size": (640, 480)}, display="lores") # Video resolution is 1440x1080 in this case
cam.configure(config)
cam.set_controls({"AfMode": controls.AfModeEnum.Manual, "LensPosition": focus})
cam.set_controls({"FrameRate": video_fps}) # video_fps was set to 24, but I also tried 30

time_now = datetime.today().strftime('%Y-%m-%d_%H-%M-%S')

file_path = f"/home/guiguig/Picamera2_Photos/RPI_{time_now}.mp4"

cam.start_and_record_video(file_path, duration=video_length, audio=True) # video_length was 6, but recorded for much longer than six seconds

Expected behaviour Audio to be recorded to the video with no errors or freezing up.

Console Output, Screenshots

[h264 @ 0x559c44f440] Thread message queue blocking; consider raising the thread_queue_size option (current value: 32)
[aac @ 0x559c46f670] Queue input is backward in time
[mp4 @ 0x559c48b610] Non-monotonous DTS in output stream 0:1; previous: 204682, current: 203062; changing to 204683. This may result in incorrect timestamps in the output file.
[mp4 @ 0x559c48b610] Non-monotonous DTS in output stream 0:1; previous: 204683, current: 204086; changing to 204684. This may result in incorrect timestamps in the output file.
[aac @ 0x559c46f670] Queue input is backward in time
[mp4 @ 0x559c48b610] Non-monotonous DTS in output stream 0:1; previous: 218422, current: 184244; changing to 218423. This may result in incorrect timestamps in the output file.
[mp4 @ 0x559c48b610] Non-monotonous DTS in output stream 0:1; previous: 218423, current: 185268; changing to 218424. This may result in incorrect timestamps in the output file.
[mp4 @ 0x559c48b610] Non-monotonous DTS in output stream 0:1; previous: 218424, current: 186292; changing to 218425. This may result in incorrect timestamps in the output file.
[mp4 @ 0x559c48b610] Non-monotonous DTS in output stream 0:1; previous: 218425, current: 187316; changing to 218426. This may result in incorrect timestamps in the output file.

Hardware : Raspberry Pi Zero 2 W, Camera Module 3, Mini USB Microphone from Digikey/Adafruit

Additional context The exact microphone I used: https://www.adafruit.com/product/3367

[goldenrubb3rs]

experiencing similar issue, also getting:

"[out_0_1 @ 0x9fdb80] 100 buffers queued in out_0_1, something may be wrong."

using a Zero 2 W with recommended Legacy 32-bit, and IQaudIO Codec Zero.

p.s. sorry I'm new to using github and all of this, just been tinkering around with this stuff and trying to teach myself so I don't really know the proper way of capturing and uploading console output. but I've just been working off the examples in the documentation for picamera2 and it seems ffmpeg is completely broken in picamera2. works fine using libcamera-vid from the terminal though.

[davidplowman]

Hi, thanks for the reports. There certainly was a problem at one point with the FfmpegOutput not stopping when there was an audio stream. Can you check what version of Picamera2 you have? Use apt list python3-picamera2 - I expect it will be 0.3.12?

If that's the case you could try checking out the 0_3_12 branch from this repository. You can either clone it directly and point your PYTHONPATH at it (this might be easier as it would let us experiment with some of the Ffmpeg parameters later), or you could get pip to install that branch directly from git. At least things shouldn't hang after that, so lets see what errors we still have then.

[davidplowman]

I've just run the following:

import time

from picamera2 import Picamera2
from picamera2.encoders import H264Encoder
from picamera2.outputs import FfmpegOutput

picam2 = Picamera2()
main = {'size': (1440, 1080)}
lores = {'size': (640, 480)}
config = picam2.create_video_configuration(main, lores=lores, display='lores')
picam2.configure(config)

encoder = H264Encoder(bitrate=10000000)
output = FfmpegOutput("test.mp4", audio=True)

picam2.start_recording(encoder, output)
time.sleep(10)
picam2.stop_recording()

and it seems to work OK on the 0_3_12 branch if I double the video and audio thread queue sizes (in picamera/picamera2/outputs/ffmpegoutput.py) to 64 an 1024 respectively. Is that a test that you could try? If confirmed that feels like a reasonable change to make.

[guighub]

Hi, sorry for the late reply.

I don't think I'm able to reinstall the microphone onto my Pi to test the code, as it's already mounted to a wooden post outdoors since I eventually decided to ditch audio recording as it wasn't entirely necessary for my project. I may purchase another camera module for my Pi 5 though to test if I have the same problem.

[davidplowman]

No problem. Well, I'll probably make that small update to the queue sizes, and also push that onto the 0_3_12 branch for Bullseye users.

[sarvasvkulpati]

Hi, just wanted to follow up on this issue.

I'm also running a Raspi Zero 2W, Bullseye, using a raspi cam v3 and a sph0645 I2s mic.

When I run audio_video_capture.py, but also the code at https://github.com/raspberrypi/picamera2/issues/902#issuecomment-1874095110, it returns the same issues:

Guessed Channel Layout for Input Stream #0.0 : stereo
[mp4 @ 0x122c370] Timestamps are unset in a packet for stream 0. This is deprecated and will stop working in the future. Fix your code to set the timestamps properly
[out_0_1 @ 0x1255520] 100 buffers queued in out_0_1, something may be wrong.
[pulse @ 0x1221e70] Thread message queue blocking; consider raising the thread_queue_size option (current value: 2048)

and remains frozen. I also tried recording a video and simultaneously recording with pyaudio, and then combining the two with ffmpeg, but that always returns no audio output. The camera and mic work independently.

Do you have any suggestions on how to fix this or what I could do?

[davidplowman]

Hi, could you try the 0_3_12 branch from this repository, there was a known issue with audio not stopping. Thanks.

[luqtas]

rpi zero 2w, pi global shutter, INMP441 mic. (works with arecord), bullseye, picamera2 0.3.12-2

[pulse @ 0x16d63b0] Thread message queue blocking; consider raisin the thread_queue_size option (current value: 512)
[out_0_1 @ 0x1704da0] 100 buffers queued in out_0_1, something may be wrong.
[h264 @ 0x16dd270] Thread message queue blocking' consider raising the thread_queue_size option (current value: 32)
default: Generic error in an external library

i tried the first solution of https://github.com/raspberrypi/picamera2/issues/467 with no avail. then i tried to modify the .py to 'alsa' and it worked. but. audio records only on one channel (it doesn't detect we are mono) and after some seconds it stops. video record is also laggy (which isn't if we are just recording video)

[davidplowman]

Have you tried the PyavOutput instead? That's a more recent (and better) way to including audio. Note that you have to set audio to True in the encoder, like this. Also check section 7.2.4 of the manual.

[nzottmann]

I do not use audio, but I get this error message also sporadically using the PyavOutput:

2025-08-08 19:58:54,266 DEBUG    - Discarded non monotonically increasing dts: 19484976901 <= 19485018555
2025-08-09 11:17:37,433 DEBUG    - Discarded non monotonically increasing dts: 74608151300 <= 74608192965
2025-08-09 15:38:40,979 DEBUG    - Discarded non monotonically increasing dts: 90271705804 <= 90271747441
2025-08-09 16:47:59,670 DEBUG    - Discarded non monotonically increasing dts: 94430383346 <= 94430425001
2025-08-09 17:15:41,891 DEBUG    - Discarded non monotonically increasing dts: 96092613450 <= 96092655105

How often this happens varies from device to device, all of them are CM4.

The only solution I found was a workaround. I created a custom encoder based on PyavOutput:

class ContinousPyavOutput(PyavOutput):
    def __init__(self, *args, **kwargs):
        self.last_timestamp = 0
        super().__init__(*args, **kwargs)

    def outputframe(self, frame, keyframe=True, timestamp=None, packet=None, audio=False):
        if timestamp <= self.last_timestamp:
            logger.debug(f"Discarded non monotonically increasing dts: {timestamp} <= {self.last_timestamp}")
            return
        
        super().outputframe(frame, keyframe, timestamp, packet, audio)
        self.last_timestamp = timestamp

Seems like there is a problem which is triggered more or less often depending on different factors like audio and something else.

Unfortunately I cannot help here with a minimal example as this happens in a bigger application and I did not have time for testing a minimal example (which would probably take a long time to wait), as I currently have this workaround. As I don't loose many frames this causes no problem currently.

I just wanted to report this to give a hint that the actual problem is possibly not only audio related.

[luqtas]

Have you tried the PyavOutput instead? That's a more recent (and better) way to including audio. Note that you have to set audio to True in the encoder, like this. Also check section 7.2.4 of the manual.

i installed Pyav with pip install av, because i though the error;

ImportError: cannot import name 'PyavOutput' from 'picamera2.outputs' (/usr/lib/python3/dist-package/picamera/outputs.__init__.py)

would be fixed but i still have the same error after installing (even tried to reboot)

same thing with apt install python3-av

[davidplowman]

Do not install using pip. Pip packages are generally much newer that the Debian/Bookworm system ones, and are often incompatible.

Uninstall the pip version and use python3-av from apt. If this does not work:

• Please confirm what OS you are using and that it's update.
• Please confirm that Picamera2 is up-to-date. I think 0.3.30 is the latest.
• Please report the actual error that stops it from being imported.
• Check that you can import av on its own, without Picamera2. Check the value of av.__file__ to be sure you're importing what you think you are importing. You should have version 12.3.0,

[luqtas]

i'm using Bullseye, which unfortunately it's my only option as otherwise i can't get my TFT display working. everything is up to date, picamera2 0.3.12-2, av version 10.x

importing av works fine

[davidplowman]

OK, I'm getting confused now. How many different threads am I trying to answer here? One issue is predominantly about the FfmpegOutput (and tangentially the ability to use PyavOutput in very old legacy builds), and the other is about using the PyavOutput on more up-to-date software where it should work correctly.

@luqtas I'm not sure how to help if you're still on Bullseye. Have you asked for help on the forums about the TFT display? (Not something I know anything about, unfortunately). The FfmpegOutput is always challenging because it pipes the video output to a separate FFmpeg process, which adds the audio itself. The FFmpeg process has to re-timestamp all the video because there's no way to give it accurate timestamps, and exactly when this happens is - or so I believe - is a large part of the problem.

@nzottmann If you're encountering a problem with the PyavOutput, perhaps we could file a separate issue for that, and then we'll just leave @luqtas 's issue live here. Please check that your software is all up to date, but it shouldn't be possible for timestamps to go backwards, so that would be something to investigate properly. Thanks!

[tvoverbeek]

@luqtas whuch TFT diaplay? Under bookworm you need a DRM (Direct Rendering Manager) driver and depending on what display software you use, some fiddling to choose the correct DRI (Direct Rendering Interface) card. Card 0 is typically the HDMI output. A TFT will probably card 1 or 2 or higher. I can use several of my SPI 320x240 TFT displays under bookworm. Also the next Debian version 13 (Trixie) is coming up soon. Then relying on Bullseye will probably become even more problematic.

[luqtas]

@tvoverbeek i'm using this one: http://www.lcdwiki.com/2.4inch_RPi_Display_For_RPi_3A+#How_to_use_with_Raspberry_Pi_OS but i also tried this PR from this lib: https://github.com/juj/fbcp-ili9341/pull/261 and both didn't worked on Bookworm

i don't mind losing the touch capabilities of the display tho