diff --git a/Dockerfile b/Dockerfile
index 34b8060d385cf837bcedc3c54be768b8edae21c8..e217e3e6b43d3310e9e12e65124b2d6accfa737c 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -11,7 +11,7 @@ RUN \
pacman -S --noconfirm --noprogressbar --quiet --needed \
git vim \
gstreamer gst-plugins-base gst-plugins-good gst-plugins-bad gst-plugins-ugly gst-libav \
- qrencode zbar ffmpeg make \
+ qrencode zbar make \
python-setuptools python-pip python-gobject gst-python python-numpy
COPY . /opt/qrlipsync
diff --git a/Makefile b/Makefile
index 67029444f4b2331b5e1166702497d4378397d8a8..ba23549f87e0932ca1817325092d0a993e495b56 100644
--- a/Makefile
+++ b/Makefile
@@ -35,7 +35,7 @@ ifeq (${VIDEO_FOUND}, 1)
@echo "${VIDEO} file not found, exiting, run with VIDEO=${VIDEO}"
@exit 1
else
- docker run -v ${CURDIR}/${VIDEO}:/opt/src/${VIDEO}:ro ${DOCKER_IMAGE_NAME} qr-lipsync-detect.py /opt/src/${VIDEO} ${ARGS}
+ docker run -v ${CURDIR}:/opt/src/ ${DOCKER_IMAGE_NAME} qr-lipsync-detect.py /opt/src/${VIDEO} ${ARGS}
endif
generate:
diff --git a/bin/qr-lipsync-detect.py b/bin/qr-lipsync-detect.py
index 4297988e2eb4975c7d3125fd31952f9d226f40c4..b8d867c8c825199833740b68afd41881bc832568 100755
--- a/bin/qr-lipsync-detect.py
+++ b/bin/qr-lipsync-detect.py
@@ -101,7 +101,7 @@ if __name__ == "__main__":
dirname = os.path.dirname(media_file)
media_prefix = os.path.splitext(os.path.basename(media_file))[0]
result_file = os.path.join(dirname, "%s_data.txt" % (media_prefix))
- d = QrLipsyncDetector.create(media_file, result_file, options, mainloop)
+ d = QrLipsyncDetector(media_file, result_file, options, mainloop)
if d:
GLib.idle_add(d.start)
try:
diff --git a/qrlipsync/detect.py b/qrlipsync/detect.py
index 87a9de71ef5953635b720de9002c7ba492e8b9ab..6be7de5547604fe4047c5aaeb203a90fd78c2f86 100644
--- a/qrlipsync/detect.py
+++ b/qrlipsync/detect.py
@@ -1,6 +1,6 @@
#!/usr/bin/env python
import os
-import shutil
+import sys
import time
import subprocess
import logging
@@ -10,9 +10,11 @@ from fractions import Fraction
import gi
gi.require_version("Gst", "1.0")
+gi.require_version("GstPbutils", "1.0")
# We don't want to use hw accel since it seems to be messing with latency
os.environ["LIBVA_DRIVER_NAME"] = "fakedriver"
from gi.repository import Gst # noqa
+from gi.repository import GstPbutils # noqa
Gst.init(None)
@@ -21,54 +23,46 @@ logger = logging.getLogger("detector")
QUEUE_OPTS = "max-size-buffers=10 max-size-bytes=0 max-size-time=0"
-def run_subprocess(cmd, filename):
- fields = cmd.split(" ")
- fields.append(filename)
- result = subprocess.check_output(fields, universal_newlines=True)
- return result
-
-
def get_media_info(media_file):
- result = None
+ uri = Gst.filename_to_uri(os.path.realpath(media_file))
try:
- ffprobe = shutil.which("ffprobe")
- except Exception:
- # python2
- from distutils.spawn import find_executable
-
- ffprobe = find_executable("ffprobe")
- if ffprobe:
- cmd = "ffprobe -v error -select_streams v -show_entries stream=width,height,avg_frame_rate,duration -of default=noprint_wrappers=1 -print_format json"
- try:
- cmd_result = run_subprocess(cmd, media_file)
- except Exception:
- cmd_result = None
- if cmd_result:
- vjres = json.loads(cmd_result)["streams"][0]
- if not vjres.get("duration"):
- cmd = "ffprobe -v error -select_streams v -show_format_entry duration -of default=noprint_wrappers=1 -print_format json"
- cmd_result = run_subprocess(cmd, media_file)
- vjres["duration"] = json.loads(cmd_result)["format"]["duration"]
- cmd = "ffprobe -v error -select_streams a -show_entries stream=sample_rate,codec_name -of default=noprint_wrappers=1 -print_format json"
- cmd_result = run_subprocess(cmd, media_file)
- ajres = json.loads(cmd_result)["streams"]
- if ajres:
- ajres = ajres[0]
- vjres["sample_rate"] = ajres["sample_rate"]
- vjres["a_codec"] = ajres["codec_name"]
- else:
- logger.error("No audio track found, cannot detect sync")
- result = vjres
+ info = GstPbutils.Discoverer.new(10 * Gst.SECOND).discover_uri(uri)
+ except gi.repository.GLib.Error as e:
+ sys.exit("Could not discover file: %s (%s)" % (media_file, e))
+
+ try:
+ vinfo = info.get_video_streams()[0]
+ except IndexError:
+ sys.exit("File contains no video stream")
+
+ fps_num = vinfo.get_framerate_num()
+ fps_denom = vinfo.get_framerate_denom()
+ if not fps_num or not fps_denom:
+ fps = None
else:
- logger.error("ffprobe is required")
+ fps = Fraction(fps_num, fps_denom)
+
+ result = {
+ 'width': vinfo.get_width(),
+ 'height': vinfo.get_height(),
+ 'framerate': fps,
+ 'duration': str(info.get_duration() / Gst.SECOND),
+ }
+
+ try:
+ ainfo = info.get_audio_streams()[0]
+ result["sample_rate"] = ainfo.get_sample_rate()
+ result["a_codec"] = GstPbutils.pb_utils_get_codec_description(ainfo.get_caps()).lower().replace("mpeg-4 aac", "aac")
+ except IndexError:
+ logger.warning("File contains no audio stream, cannot detect sync")
return result
class QrLipsyncDetector:
- def __init__(self, media_file, result_file, options, mainloop, media_info):
+ def __init__(self, media_file, result_file, options, mainloop):
self.analyze_returncode = None
self.options = options
- self.media_info = media_info
+ self.media_info = get_media_info(media_file)
self._samplerate = int(self.media_info.get("sample_rate", 0))
self._media_duration = float(self.media_info["duration"])
self.mainloop = mainloop
@@ -87,20 +81,20 @@ class QrLipsyncDetector:
self._tick_count = 0
spectrum_interval_ms = 3
self.spectrum_interval_ns = spectrum_interval_ms * Gst.MSECOND
- framerate = self.media_info.get("avg_frame_rate")
- if framerate is not None:
- # assume audio ticks are at least 1 video frame long
- self.framerate = Fraction(self.media_info["avg_frame_rate"])
- frame_dur_ms = float(1000 / self.framerate)
- else:
- # assume 60 fps
- frame_dur_ms = 1000 / 60
+ framerate = self.media_info.get('framerate')
+ if framerate is None:
+ if options.expected_beep_duration:
+ framerate = 1000 / options.expected_beep_duration
+ logger.warning(f'Unable to guess framerate, using --expected-beep-duration to guess framerate: {framerate}')
+ else:
+ framerate = Fraction(60, 1)
+ logger.warning('Unable to guess framerate, assuming {framerate} until we can extract the real value from the first qrcode')
+ self.framerate = framerate
- if options.expected_beep_duration:
- self.ticks_count_threshold = int(options.expected_beep_duration / spectrum_interval_ms)
- else:
- self.ticks_count_threshold = int(frame_dur_ms / spectrum_interval_ms)
+ # assume audio ticks are at least 1 video frame long
+ frame_dur_ms = float(1000 / self.framerate)
+ self.ticks_count_threshold = int(frame_dur_ms / spectrum_interval_ms)
# FIXME: fdk adds 2048 samples of priming samples (silence) which adds 42ms of latency
# aacenc adds 1024 samples (21ms)
@@ -137,14 +131,6 @@ class QrLipsyncDetector:
self.pipeline_str = self.get_pipeline(self._uri_media_file)
self.pipeline = Gst.parse_launch(self.pipeline_str)
- @classmethod
- def create(qrlipsyncdetector, media_file, result_file, options, mainloop):
- result = None
- media_info = get_media_info(media_file)
- if media_info:
- result = qrlipsyncdetector(media_file, result_file, options, mainloop, media_info)
- return result
-
def exit(self):
self.pipeline.set_state(Gst.State.NULL)
self.mainloop.quit()
@@ -304,7 +290,11 @@ class QrLipsyncDetector:
real_framerate = float(Fraction(qrcode["FRAMERATE"]))
beep_dur = int(1000 / real_framerate)
if real_framerate != self.framerate:
- logger.warning(f"Input file framerate ({self.framerate}) differs from original sample framerate ({real_framerate}), you should run this with --expected-beep-duration {beep_dur} or beeps won't be detected")
+ logger.warning(f"Detected real framerate {real_framerate} (different from assumed framerate {self.framerate}, updating values. You should run this with --expected-beep-duration {beep_dur} or (some) beeps may not be detected")
+ self.framerate = real_framerate
+ frame_dur_ms = 1000 / real_framerate
+ spectrum_interval_ms = self.spectrum_interval_ns * 1000
+ self.ticks_count_threshold = int(frame_dur_ms / spectrum_interval_ms)
qrcode["ELEMENTNAME"] = elt_name
qrcode["VIDEOTIMESTAMP"] = timestamp
if qrcode.get("TICKFREQ"):
@@ -372,6 +362,12 @@ class QrLipsyncDetector:
self._tick_count += 1
self.write_line(json.dumps(result))
+ def run_subprocess(self, cmd, filename):
+ fields = cmd.split(" ")
+ fields.append(filename)
+ result = subprocess.check_output(fields, universal_newlines=True)
+ return result
+
def disconnect_probes(self):
logger.debug("Disconnecting probes")
if self._audio_fakesink_pad: