Add audio sample count calculation function while maintain precision of the timestamp

[drunkenQCat]

I am a programmer working on sound, my work always needs to transfer SMPTE timecode to time stamp, and times them to 44100 or 48000.
我是音频技术，对SMPTE时间码转时间戳有需求。日常工作中，时间戳的结果要乘以44100或者48000。

But after running code below, the result appears so wired.
但是下面这段代码得出了预料之外的结果：

from dftt_timecode import DfttTimecode

a = DfttTimecode('01:00:00:13', 'smpte', fps=24, drop_frame=False, strict=True)
print(f"the timestamp of 01:00:00:13 is :  {a.timestamp}")
print(f"the sample count at 44100 is :  { a.timestamp * 44100 }")
print(f"the sample count at 48000 is :  { a.timestamp * 48000 }")

print("================================================")

a = DfttTimecode('01:00:00:14', 'smpte', fps=24, drop_frame=False, strict=True)
print(f"the timestamp of 01:00:00:14 is :  {a.timestamp}")
print(f"the sample count at 44100 is :  { a.timestamp * 44100 }")
print(f"the sample count at 48000 is :  { a.timestamp * 48000 }")

the timestamp of 01:00:00:13 is :  3600.54167
the sample count at 44100 is :  158783887.647
the sample count at 48000 is :  172826000.16
================================================
the timestamp of 01:00:00:14 is :  3600.58333
the sample count at 44100 is :  158785724.853
the sample count at 48000 is :  172827999.84

I think the problem may happens in data type tranferring or something else, but I haven't inspect the source code yet. Or maybe the maintainer has any consideration on it?
我觉得可能是数据类型转换时候出了问题，不过我还没有查到是源代码哪部分。或者这是作者设计上的考量吗？

[OwenYou]

Hello! It's glad to know that this library is being used by the industry.

The problem you encountered is caused by multiplying a timestamp by a large value. timestamp is not designed to be used in such scenario, the value is already rounded before returning. So, instead, I recommend you to use precise_timestamp for this scenario.

a = DfttTimecode('01:00:00:13', 'smpte', fps=24, drop_frame=False, strict=True)
print(f"the timestamp of 01:00:00:13 is :  {a.timestamp}")
print(f"the sample count at 44100 is :  {float(a.precise_timestamp * 44100):.2f}")
print(f"the sample count at 48000 is :  {float(a.precise_timestamp * 48000):.2f}")

print("================================================")

a = DfttTimecode('01:00:00:14', 'smpte', fps=24, drop_frame=False, strict=True)
print(f"the timestamp of 01:00:00:14 is :  {a.timestamp}")
print(f"the sample count at 44100 is :  {float(a.precise_timestamp * 44100):.2f}")
print(f"the sample count at 48000 is :  {float(a.precise_timestamp * 48000):.2f}")

the timestamp of 01:00:00:13 is :  3600.54167
the sample count at 44100 is :  158783887.50
the sample count at 48000 is :  172826000.00
================================================
the timestamp of 01:00:00:14 is :  3600.58333
the sample count at 44100 is :  158785725.00
the sample count at 48000 is :  172828000.00

You can see that the code is giving the right result. 1 frame at 24 fps corresponds to 1837.5 samples at 44.1 kHz, or 2000 samples at 48 kHz.

But still, this is not quite an elegant way of use. So I am considering of packing this function(give out the audio sample counts of a tc instance under certain sample rate) into a class method.

Hope this resolves your problem!

Best regards.

[OwenYou]

@WheheoHu We should consider add a class method to make such operations easier. Maybe the implementation should like:

# class method
get_audio_sample_counts(self, sample_rate):
    return math.floor(self.__precise_time*sample_rate)
# floor to avoid 'ghost' sample

[drunkenQCat]

Thank you for your response! I will use this method in my project moving forward. I look forward to seeing this package being applied more widely in the future.