Update icon and title audio classification

[st-tuanmai]

No description provided.

[PaulTR]

Can you verify the classification happening in the app? It's picking up a lot of random things that aren't happening, and not picking up things like "speech" or "whistling" when that is happening. Thanks.

[st-tuanmai]

Can you verify the classification happening in the app? It's picking up a lot of random things that aren't happening, and not picking up things like "speech" or "whistling" when that is happening. Thanks.

I will check it now

[st-tuanmai]

I updated the branch to change TensorFlowLiteTaskAudio to TensorFlowLiteSwift.
Please help me check again.
Thanks you.

[PaulTR]

I am still seeing wrong results consistently with this. Just whistling into the phone and not getting 'whistling'.

[st-tuanmai]

Hi @PaulTR
I checked the model label file and the speech_commands label doesn't have whistling and speech. Please help me check.

labels:
background
down
go
left
off
on
right
stop
up

[PaulTR]

It shouldn't be the speech one with whistling - it's the other one that isn't registering it.

…

On Mon, Oct 28, 2024, 8:25 PM Tuan Mai A. ***@***.***> wrote: Hi @PaulTR <https://github.com/PaulTR> I checked the model label file and the speech_commands label doesn't have whistling and speech. Please help me check. labels: background down go left off on right stop up — Reply to this email directly, view it on GitHub <#38 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAR2C37VZLKMMS7DO4GCTYTZ53W2JAVCNFSM6AAAAABPRGAHCOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDINBTGAZTIOBRGA> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[PaulTR]

You can see the MediaPipe sample to see what it should look like. They use the same models.

…

On Mon, Oct 28, 2024, 8:26 PM Paul Trebilcox-Ruiz ***@***.***> wrote: It shouldn't be the speech one with whistling - it's the other one that isn't registering it. On Mon, Oct 28, 2024, 8:25 PM Tuan Mai A. ***@***.***> wrote: > Hi @PaulTR <https://github.com/PaulTR> > I checked the model label file and the speech_commands label doesn't have > whistling and speech. Please help me check. > > labels: > background > down > go > left > off > on > right > stop > up > > — > Reply to this email directly, view it on GitHub > <#38 (comment)>, > or unsubscribe > <https://github.com/notifications/unsubscribe-auth/AAR2C37VZLKMMS7DO4GCTYTZ53W2JAVCNFSM6AAAAABPRGAHCOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDINBTGAZTIOBRGA> > . > You are receiving this because you were mentioned.Message ID: > ***@***.***> >

[st-tuanmai]

I am using label.txt file extracted from tflite file using python code, I think label is added wrongly to metadata.
vocal = zipfile.ZipFile('./demo/speech_commands.tflite').extractall('label')

[PaulTR]

Not comparing the speech commands model - can you verify this all with the standard audio classification model (the one with 'music', 'whispering', 'whistling', etc.)? Thanks.

[st-tuanmai]

I checked several sources and it seems the models used are different so the labels are different too.
https://github.com/tensorflow/examples/blob/master/lite/examples/sound_classification/ios/SoundClassification/Model/labels.txt
https://www.tensorflow.org/datasets/catalog/speech_commands
https://research.google/blog/launching-the-speech-commands-dataset/

[PaulTR]

OK again we're not looking at the speech models for this issue. Only looking at the sound classification model that does whistling/whispering/music/etc. the label that shows up is not matching the sound.

Delete the speech model if that's adding a complication, it doesn't matter for the sample. We need the first general sound classification model to work correctly.

[PaulTR]

You can compare the iOS sample to the Android sample. They're using this model: https://storage.googleapis.com/ai-edge/interpreter-samples/audio_classification/android/yamnet.tflite (I copied the same one into the ios folder) and returning the correct results.

[st-tuanmai]

Hi Paul,
iOS using this yamnet model:
https://storage.googleapis.com/ai-edge/interpreter-samples/audio_classification/ios/yamnet.tflite

[PaulTR]

Right, and they should be the same model.

…

On Tue, Nov 5, 2024, 7:28 PM Tuan Mai A. ***@***.***> wrote: Hi Paul, iOS using this yamnet model: https://storage.googleapis.com/ai-edge/interpreter-samples/audio_classification/ios/yamnet.tflite — Reply to this email directly, view it on GitHub <#38 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAR2C36QCC2E372XTHZJL3DZ7F5ELAVCNFSM6AAAAABPRGAHCOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDINJYGYYDINZYHE> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[st-tuanmai]

@PaulTR I removed the speech commands model, I think we can use only yamnet model now.