data sets

[wwyl2000]

Dear Author,
Thanks for sharing your package. In your example to generate the data set, "fire" has 2 parts of data, positive and negative. What is the positive data? Was it pre-recorded? Also, if i have a new word to detect, for example, hakunamatata, how to obtain the datasets?

Thanks,
WWY

[ljj7975]

positive refers to audios with target keyword (fire).
negative are the audios without target keyword (fire).

training on negative set helps decreasing false positive rate.

Unfortunately, there isn't a good way of generating a dataset for custom wakeword.
If it is made up of common word such as hey, hi, cat. Data generation using Mozilla Dataset should work.

However, generating a dataset for non-standard word such as hakunamatata is not yet supported.

[wwyl2000]

Hi ljj7975,
Many thanks for your informaion.
Best,
wwy

[Ruiqin-Huang]

Howl is a frame-level wake word detection system. When training for a single English wake word like "fire", should the training/development/test sets:

Option A: Contain only audio samples with the exact wake word?
Example:

• sample_1's transcription: "fire"
• sample_2's transcription: "fire"

OR

Option B: Contain sentences that include the wake word?
Example:

• sample_1's transcription: "fire in the sky"
• sample_2's transcription: "The building was on fire, with flames engulfing the upper floors"