Recherche avancée

Sur d’autres sites (7452)

• Split h.264 stream into multiple parts in python

  31 janvier 2023, par BillPlayz

  My objective is to split an h.264 stream into multiple parts, meaning while reading the stream from a pipe i would like to save it into x second long packages (in my case 10).

  


  I am using a libcamera-vid subprocess on my Raspberry Pi that outputs the h.264 stream into stdout.
  
Might be irrelevant, depends : libcamera-vid outputs a message every frame and I am able to locate it at isFrameStopLine
To convert the stream, I use an ffmpeg subprocess, as you can see in the code below.

  


  Imagine it like that :
  
Stream is running...
  
- Start recording to a file
  
- Sleep x seconds
  
- Finish recording to file
  
- Start recording a new file
  
- Sleep x seconds
  
- Finish recording the new file
  
- and so on...

  


  Here is my current code, however upon running the first export succeeds, and after the second or third the ffmpeg-subprocess is terminating with the error :
  
pipe:: Invalid data found when processing input
  
And shortly after, the python process, because of the ffmpeg termination i believe.
Traceback (most recent call last):   File "/home/survpi-camera/main.py", line 56, in <module>     processStreamLine(readData)   File "/home/survpi-camera/main.py", line 16, in processStreamLine     streamInfo["process"].stdin.write(data) BrokenPipeError: [Errno 32] Broken pipe</module>

  


  recentStreamProcesses = []
streamInfo = {
    "lastStreamStart": -1,
    "process": None
}

def processStreamLine(data):
    isInfoLine = ((data.startswith(b"[") and (b"INFO" in data)) or (data == b"Preview window unavailable"))
    isFrameStopLine = (data.startswith(b"#") and (b" fps) exp" in data))
    if ((not isInfoLine) and (not isFrameStopLine)):
        streamInfo["process"].stdin.write(data)
    
    if (isFrameStopLine):
        if (time.time() - streamInfo["lastStreamStart"] >= 10):
            print("10 seconds passed, exporting...")
            exportStream()
            createNewStream()

def createNewStream():
    streamInfo["lastStreamStart"] = time.time()
    streamInfo["process"] = subprocess.Popen([
        "ffmpeg",
        "-r","30",
        "-i","-",
        "-c","copy",("/home/survpi-camera/" + str(round(time.time())) + ".mp4")
    ],stdin=subprocess.PIPE,stderr=subprocess.STDOUT)
    print("Created new streamProcess.")

def exportStream():
    print("Exporting...")
    streamInfo["process"].stdin.close()
    recentStreamProcesses.append(streamInfo["process"])


cameraProcess = subprocess.Popen([
    "libcamera-vid",
    "-t","0",
    "--width","1920",
    "--height","1080",
    "--codec","h264",
    "--inline",
    "--listen",
    "-o","-"
],stdout=subprocess.PIPE,stdin=subprocess.PIPE,stderr=subprocess.STDOUT)

createNewStream()

while True:
    readData = cameraProcess.stdout.readline()
    
    processStreamLine(readData)


  


  Thank you in advance !

  


• ffmpeg hwo to choose the best bit rate for HLS streaming ?

  7 février 2023, par Josef Kranz

  I'm building a video streaming white label product, but I've run into the following scenario where I'm not sure what might be the optimal way to get the best video quality in return. At the moment I'm using crf based encoding, which is from a streaming based point of view a very bad idea. From a quality point of view, it will definitely give me the best quality and also the best efficiency. Why is crf based encoding bad for streaming ? First you have a non-fixed file size, meaning your bit rate might vary between 20 kbps and 20 Mbit/s depending on the codec and picture motion of the current frames ...
When you have such heavily varying stream according to the bit rate, automatic quality selection at the player might not function correctly, and the player automatically switches from 1080p to 480p for no reason (using VideoJS here).

  


  To fix this issue, it's fine if I set minrate, maxrate with ffmpeg but this will also come at the cost that some frames might look pixelated, which I absolutely do not want.

  


  Currently, my encoding command looks something like this :

  


  /usr/bin/ffmpeg -i "/tmp/VODProcessing/Tester 2160p UHD-HDR.mp4" -map 0:0 -c:v libsvtav1 -crf 19 -vf zscale=width=3840:height=2160 -svtav1-params "profile=0:enable-force-key-frames=1:superres-mode=1:enable-tf=0:tune=0:enable-overlays=1:scd=1:scm=2:enable-mfmv=1:enable-cdef=1:enable-dlf=1:fast-decode=1:color-primaries=9:transfer-characteristics=16:matrix-coefficients=9:input-depth=10:mastering-display=G(0.265,0.69)B(0.15,0.06)R(0.68,0.32)WP(0.3127,0.329)L(4000.0,0.005):content-light=368,226:enable-hdr=1:color-range=1" -pix_fmt yuv420p10le -color_trc smpte2084 -color_primaries bt2020 -colorspace bt2020nc -chroma_sample_location:v topleft -color_range:v pc -max_muxing_queue_size 1024 -preset 7 -bf 0 -force_key_frames "expr:gte(t,n_forced*4.004)" -keyint_min 48 -sc_threshold 0 -use_timeline 1 -use_template 1 -map_metadata -1 -map_chapters -1 -f hls -seg_duration 4.004 -hls_time 4.004 -streaming 1 -hls_list_size 0 -hls_segment_filename "/tmp/VODProcessing/output/Tester 2160p UHD-HDR/v-av01-2160p-av01.0.12M.10_PQ/f-%04d.m4s"  -hls_fmp4_init_filename "init-v-av01-2160p-av01.0.12M.10_PQ.m4s"  -hls_segment_type fmp4 -hls_playlist_type vod -movflags frag_keyframe+frag_every_frame+write_colr+prefer_icc+skip_trailer+faststart -hls_flags independent_segments -strict experimental "/tmp/VODProcessing/output/Tester 2160p UHD-HDR/v-av01-2160p-av01.0.12M.10_PQ/master.m3u8"


  


  Which will form a stream made out of independent m4s segments. As you can see, I'm using libsvtav1 in crf mode to output the result in AV1.

  


  Now my question is, how can I have the same nice quality output as with the crf mode while having a static/fixed bit rate ? Will 2 pass encoding solve this problem by distributing the pixels or data rate differently ?

  


  Thanks in advance

  


• How to use audio frame after decode mp3 file using pyav, ffmpeg, python

  2 janvier 2021, par Long Tran Dai

  I am using using python with pyav, ffmpeg to decode mp3 in the memory. I know there are some other way to do it, like pipe ffmpeg command. However, I would like to explore pyav and ffmpeg API. So I have the following code. It works but the sound is very noisy, although hearable :

  


  import numpy as np&#xA;import av # to convert mp3 to wav using ffmpeg&#xA;import pyaudio # to play music&#xA;&#xA;mp3_path = &#x27;D:/MyProg/python/SauTimThiepHong.mp3&#x27;&#xA;&#xA;def decodeStream(mp3_path):&#xA;  # Run NOT OK&#xA;  &#xA;  container = av.open(mp3_path)&#xA;  stream = next(s for s in container.streams if s.type == &#x27;audio&#x27;)&#xA;  frame_count = 0&#xA;  data = bytearray()&#xA;  for packet in container.demux(stream):&#xA;    # <class>&#xA;    # We need to skip the "flushing" packets that `demux` generates.&#xA;    #if frame_count == 5000 : break         &#xA;    if packet.dts is None:&#xA;        continue&#xA;    for frame in packet.decode():   &#xA;        #&#xA;        # type(frame) : <class>&#xA;        #frame.samples = 1152 : 1152 diem du lieu : Number of audio samples (per channel)&#xA;        # moi frame co size = 1152 (diem) * 2 (channels) * 4 (bytes / diem) = 9216 bytes&#xA;        # 11021 frames&#xA;        #arr = frame.to_ndarray() # arr.nbytes = 9216&#xA;&#xA;        #channels = []  &#xA;        channels = frame.to_ndarray().astype("float16")&#xA;        #for plane in frame.planes:&#xA;            #channels.append(plane.to_bytes()) #plane has 4 bytes / sample, but audio has only 2 bytes&#xA;        #    channels.append(np.frombuffer(plane, dtype=np.single).astype("float16"))&#xA;            #channels.append(np.frombuffer(plane, dtype=np.single)) # kieu np.single co 4 bytes&#xA;        if not frame.is_corrupt:&#xA;            #data.extend(np.frombuffer(frame.planes[0], dtype=np.single).astype("float16")) # 1 channel: noisy&#xA;            # type(planes) : <class>&#xA;            frame_count &#x2B;= 1&#xA;            #print( &#x27;>>>> %04d&#x27; % frame_count, frame)   &#xA;            #if frame_count == 5000 : break     &#xA;            # mix channels:&#xA;            for i in range(frame.samples):                &#xA;                for ch in channels: # dec_ctx->channels&#xA;                    data.extend(ch[i]) #noisy&#xA;                    #fwrite(frame->data[ch] &#x2B; data_size*i, 1, data_size, outfile)&#xA;  return bytes(data)&#xA;</class></class></class>

  


  I use pipe ffmpeg to get decoded data to compare and find they are different :

  


  def RunFFMPEG(mp3_path, target_fs = "44100"):
    # Run OK
    import subprocess
    # init command
    ffmpeg_command = ["ffmpeg", "-i", mp3_path,
                   "-ab", "128k", "-acodec", "pcm_s16le", "-ac", "0", "-ar", target_fs, "-map",
                   "0:a", "-map_metadata", "-1", "-sn", "-vn", "-y",
                   "-f", "wav", "pipe:1"]
    # excute ffmpeg command
    pipe = subprocess.run(ffmpeg_command, stdout=subprocess.PIPE, stderr=subprocess.PIPE, bufsize= 10**8)
    # debug
    #print(pipe.stdout, pipe.stderr)
    # read signal as numpy array and assign sampling rate
    #audio_np = np.frombuffer(buffer=pipe.stdout, dtype=np.uint16, offset=44)
    #audio_np = np.frombuffer(buffer=pipe.stdout, dtype=np.uint16)
    #sig, fs  = audio_np, target_fs
    #return audio_np
    return pipe.stdout[78:]     


  


  Then I use pyaudio to play data and find it very noisy

  


  p = pyaudio.PyAudio()
streamOut = p.open(format=pyaudio.paInt16, channels=2, rate= 44100, output=True)
#streamOut = p.open(format=pyaudio.paInt16, channels=1, rate= 44100, output=True)

mydata = decodeStream(mp3_path)
print("bytes of mydata = ", len(mydata))
#print("bytes of mydata = ", mydata.nbytes)

ffMpegdata = RunFFMPEG(mp3_path)
print("bytes of ffMpegdata = ", len(ffMpegdata)) 
#print("bytes of ffMpegdata = ", ffMpegdata.nbytes)

minlen = min(len(mydata), len(ffMpegdata))
print("mydata == ffMpegdata", mydata[:minlen] == ffMpegdata[:minlen]) # ffMpegdata.tobytes()[:minlen] )

#bytes of mydata =  50784768
#bytes of ffMpegdata =  50784768
#mydata == ffMpegdata False

streamOut.write(mydata)
streamOut.write(ffMpegdata)
streamOut.stop_stream()
streamOut.close()
p.terminate()


  


  Please help me to understand decoded frame of pyav api (after for frame in packet.decode() :). Should it be processed more ? or I have some error ?

  


  It makes me crazy for 3 days. I could not guess where to go.

  


  Thank you very much.