Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{203840,
        author = {hemalatha B and Dr Krishna Kumar P R},
        title = {End to end speech emotion recognition using CNN and Spectrogram with gender prediction},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {1},
        pages = {485-498},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=203840},
        abstract = {Figuring out feelings through voice has become a big deal in AI and machine learning circles. This effort aims to build a smart setup that catches emotions in spoken words - at the same time guessing if the speaker is male or female - by relying on convolutional neural networks alongside sound visuals called spectrograms. Instead of treating sounds as raw data, it turns them into images first. Working behind the scenes, signal processing joins forces with deep learning models so emotional tones can be sorted more precisely. Accuracy steps up when both types of tech run together rather than alone.
From open-source voice collections showing feelings like joy, sorrow, rage, terror, shock, dislike, and no emotion at all - sound files are gathered here. Before anything else happens, background hums get stripped out; clarity takes priority during cleanup steps. Normalization adjusts volume peaks, while quiet gaps vanish so every clip holds useful information only. Changing playback speed uniformly helps align inputs across different sources despite original variations. Once cleaned, key characteristics emerge through image-like maps built from sound waves over time. Mel-scale pictures of frequency shifts reveal how voices twist when emotions change unexpectedly. MFCC values pull out subtle vocal textures machines later recognize as distinct mood markers. These visuals act much like snapshots that highlight rhythm, pitch bends, and intensity bursts uniquely tied to each feeling. Convolution networks study those images just as they would photos - spotting tiny clues hidden in wave shapes. Emotion-linked habits within speech slowly become visible after repeated scanning by learning systems.
From those captured spectrogram pictures, a special kind of neural network takes over - its job is spotting patterns tied to emotion and voice differences between genders. Instead of hand-picking features, the system learns them on its own through layered processing steps. Layer after layer, it sharpens what it recognizes, stacking convolutions with down sampling and fully connected stages. Performance gets checked thoroughly, using measures like correct guess rates, consistency scores, detection completeness, balance indicators, plus detailed error mapping. Each step feeds into how well the whole setup classifies voices.
Accuracy matters most when spotting feelings and voices in sound, yet speed cannot slow down. One idea runs through every use: machines that listen better start here. Virtual helpers answer questions, sure - but now they might sense stress too. Customer service tools shift quietly, adapting to tone instead of just words. Mental health check ins? They gain subtle clues from how someone speaks, not only what is said. Call centers grow smarter by catching frustration before it spreads. Chatbots react differently if sadness shows up mid-sentence. Security setups watch for vocal tension like a guard watching shadows. Even classrooms change when software adjusts lessons based on student mood shifts. Deep networks meet audio filters somewhere between labs and daily life. Human talk gets broken into pieces, then rebuilt as meaning machines grasp. Interaction evolves without flashy promises - simply closer mimicry of natural listening. Understanding grows not from rules, but patterns found in thousands of spoken moments.},
        keywords = {},
        month = {June},
        }