devicehive / devicehive-audio-analysis Goto Github PK

I'm testing with an 18 second long file WAV and the predictions array is empty - is there a recommended minimum duration? Here is the avprobe output

Input #0, wav, from 'test.wav':
  Metadata:
    encoder         : Lavf56.1.0
  Duration: 00:00:18.25, bitrate: 256 kb/s
    Stream #0.0: Audio: pcm_s16le, 16000 Hz, 1 channels, s16, 256 kb/s

1. Is your result close to what published by the audioset paper? How do you define hit rate? Is it accuracy?
2. Is the feature you generated using vggish model close to their released feature? It seems very different. 3. They have quantized 8-bit feature and you used floating number feature? Is your model trained with their 8-bit feature?

Hello,

I am getting this error while trying to run the code via command line on Ubuntu. I use the command python3 parse_file.py Recording_5.wav

Here is the Traceback:

Traceback (most recent call last):
File "parse_file.py", line 39, in
process_file(**vars(args))
File "parse_file.py", line 31, in process_file
with WavProcessor() as proc:
File "/home/akshay/devicehive-audio-analysis/audio/processor.py", line 40, in init
pca_params = np.load(params.VGGISH_PCA_PARAMS)
File "/home/akshay/.local/lib/python3.6/site-packages/numpy/lib/npyio.py", line 428, in load
fid = open(os_fspath(file), "rb")
FileNotFoundError: [Errno 2] No such file or directory: 'models/vggish_pca_params.npz'

For both variants (from wav and mic recording) that I try - same error:
Traceback (most recent call last): File "parse_file.py", line 39, in <module> process_file(**vars(args)) File "parse_file.py", line 31, in process_file with WavProcessor() as proc: File "/mnt/Data/project/devicehive-audio-analysis/audio/processor.py", line 45, in __init__ self._init_youtube() File "/mnt/Data/project/devicehive-audio-analysis/audio/processor.py", line 74, in _init_youtube youtube8m.model.load_model(sess, params.YOUTUBE_CHECKPOINT_FILE) File "/mnt/Data/project/devicehive-audio-analysis/audio/utils/youtube8m/model.py", line 43, in load_model set_up_init_ops(tf.get_collection_ref(tf.GraphKeys.LOCAL_VARIABLES)) File "/mnt/Data/project/devicehive-audio-analysis/audio/utils/youtube8m/model.py", line 26, in set_up_init_ops if "train_input" in variable.name: AttributeError: 'str' object has no attribute 'name'
Model was downloaded from your link.

Hi,
First of all, great work! I would like to get exact time information of the previously trained objects rather than the percentage of them. Is it possible if I tweak the current project?

hi,dear
I found that you get the features with postprocess, here
if I must postprocess ?

thx

Hi @igor-panteleev, great job. I want to train my own model for two or three particular classes. I'm planning the following training pipeline.

128 dim embedding --> a classifier --> classes

Though I've seen the Google Audio set data is provided in 128 dim tf.records, I couldn't find it in a downloadable form in the site. I found a frame by frame tensorflow.SequenceExample file of 2.4 GB. But is that the same data you've used? Please help me on this.

On evaluation, the pipeline will be

WAV format --> VGGish --> 128 dim embedding --> a classifier --> prediction label

Do I miss something?

hi, could you share a well pre_trained youtube8m-model? The existing you provided in your project youtube_model.ckpt.data-00000-of-00001 is not very accurate when run the demo server.

Thank you for sharing such a audio analysis project!I wonder if I can use it on the Windows? Or it's limited in Ubuntu system?

where is the dh_webconfig?

Good night.

I want to create my own dataset with my own labels. is it possible for this repository?

Thanks

how to deploy this application in heroku ?

Hello! Great work!

I'm new in audio analysis and tensorflow...

How i change the dataset?

I want to use a offline dataset in my computer(Example: Urbansound). I can train this dataset and use it in place of youtube-8M?

Thanks!

-e git://github.com/devicehive/devicehive-python-webconfig.git@a792db1babcedb5baa68ec6ba6ebcf0041f20469#egg=devicehive_webconfig

release file not available , can you please update the repository with the latest codes