[ABSTRAK
Pesatnya perkembangan Deep Learning akhir-akhir ini juga menyentuh ASRberbasis HMM, sehingga memunculkan teknik hibrid HMM-ANN. Salah satuteknik Deep Learning yang cukup menjanjikan adalah penggunaan arsitekturCNN. CNN yang memiliki kemampuan mendeteksi local correlation sesuaiuntuk digunakan pada data spectrum suara. Spectrogram memiliki karakteristiklocal correlation yang nampak secara visual. Penelitian ini adalah eksperimenpenggunaan spectrogram sebagai fitur untuk HMM-CNN untuk meningkatkankinerja ASR berbasis HMM. Penelitian menyimpulkan spectogram dapatdigunakan sebagai fitur untuk HMM-CNN untuk meningkatkan kinerja ASRberbasis HMM.

---

ABSTRACT
The latest surge in Deep Learning affecting HMM based ASR, which give birth tohybrid HMM-ANN technique. One of the promising Deep Learning technique isthe implementation of CNN architecture. The ability of CNN to detect localcorrelation make it suitable to be used for speech spectral data. Spectrogram as aspeech spectral data has local correlation characteristic which is visuallyobservable. This research is an experiment to use spectrogram as a feature forHMM-CNN to add to the performance of HMM based ASR. This research foundthat spectrogram is indeed can be used as a feature for CNN to add to theperformance of HMM based ASR., The latest surge in Deep Learning affecting HMM based ASR, which give birth tohybrid HMM-ANN technique. One of the promising Deep Learning technique isthe implementation of CNN architecture. The ability of CNN to detect localcorrelation make it suitable to be used for speech spectral data. Spectrogram as aspeech spectral data has local correlation characteristic which is visuallyobservable. This research is an experiment to use spectrogram as a feature forHMM-CNN to add to the performance of HMM based ASR. This research foundthat spectrogram is indeed can be used as a feature for CNN to add to theperformance of HMM based ASR.]