--- title: "Speech recognition and generation" subject: "Multimedia System" chapter: 2 key: "semester_8/multimedia_system:speech-recognition-generation" source: https://www.bctengineer.com/q/semester_8/multimedia_system/speech-recognition-generation site: BCT Engineering --- # Speech recognition and generation Multimedia System · Chapter 2: Sound / Audio System - Asked in: 2075 Bhadra, 5 marks - Video: [Brief survey of speech recognition and generation](https://www.youtube.com/watch?v=dpAdCglwqUo) — VR Study With Target - Answer checked by Claude Opus 5.5 (high effort) on 2026-10-14 - Page: https://www.bctengineer.com/q/semester_8/multimedia_system/speech-recognition-generation - Whole chapter: https://www.bctengineer.com/md/semester_8/multimedia_system/chapter-2 ## Question Explain in detail the process of speech recognition and generation. ## Short answer Speech generation converts text to speech using concatenation or formant synthesis, with text normalisation, phonemes, prosody and waveform generation. Speech recognition samples and extracts features, matches patterns or models, then applies linguistic rules. It can be speaker-dependent or independent, isolated or continuous. ## Answer **Speech generation (synthesis)** produces spoken output from text or stored data. **Speech recognition** converts a spoken signal into text or commands. The two are inverse processes and are the basis of voice user interfaces. #### Speech generation 1. **Time-dependent sound concatenation.** Short recorded speech units (phonemes, diphones, syllables or words) are stored and joined in sequence. Joins are smoothed so there are no clicks. Quality is natural but needs a large database and care at the joints (co-articulation). 2. **Frequency-dependent sound concatenation (formant synthesis).** The vocal tract is modelled by resonance filters. Parameters such as fundamental frequency (pitch) and formants (resonances) are generated for each sound and fed to a filter. It needs little storage but sounds more artificial. ```text Text -> Text analysis -> Phoneme/prosody -> Sound (numbers, (pitch, duration) synthesis abbreviations) | Speech ``` Steps of text-to-speech: text normalisation, conversion to phonemes, adding prosody (stress, intonation, rhythm), then generating the waveform. #### Speech recognition ```text Speech -> Feature -> Compare with -> Recognised signal extraction reference word/text (spectrum) patterns + language rules ``` 1. **Sampling and pre-processing:** the signal is digitised and noise is reduced. 2. **Feature extraction:** the signal is cut into short frames (about 20 ms) and features such as spectrum or cepstral coefficients are computed. 3. **Pattern matching:** features are compared with stored reference patterns or statistical models (templates, Hidden Markov Models, neural networks) to find the closest phoneme or word. 4. **Linguistic analysis:** grammar and word probabilities (syntax and semantics) are used to choose the best sentence. Recognition may be **speaker-dependent** (trained to one voice, more accurate) or **speaker-independent**, and **isolated word** or **continuous speech**. Difficulties are noise, accents, speed of speaking and similar-sounding words. ## Check yourself The answer's key points as multiple-choice questions, with the correct option and why. 1. What is the main idea of frequency-dependent (formant) synthesis? A. Joining stored recorded phonemes in sequence B. Recording whole sentences C. Modelling the vocal tract with resonance filters driven by pitch and formant parameters D. Matching features to templates Answer: C. Formant synthesis uses filters for vocal tract resonance; it needs little storage but sounds artificial. (Key point: Formant synthesis) 2. What is the correct order of steps in text-to-speech? A. Waveform generation, phonemes, prosody, normalisation B. Phonemes, normalisation, waveform, prosody C. Prosody, waveform, normalisation, phonemes D. Text normalisation, phonemes, prosody, waveform generation Answer: D. Text is normalised, converted to phonemes, given prosody, then the waveform is generated. (Key point: Text-to-speech steps) 3. In speech recognition, what is done during feature extraction? A. The signal is cut into short frames (about 20 ms) and spectrum or cepstral features computed B. Grammar rules choose the best sentence C. The text is spoken aloud D. Stored phonemes are joined Answer: A. Feature extraction splits the signal into about 20 ms frames and computes features such as the spectrum. (Key point: Recognition steps) 4. Which statement about speaker-dependent recognition is correct? A. It works for any voice without training B. It is trained to one voice and is more accurate C. It is less accurate than speaker-independent D. It can only recognise continuous speech Answer: B. Speaker-dependent systems are trained for one voice and are more accurate. (Key point: Speaker dependence) ## Related in Multimedia System - [Image recognition steps](https://www.bctengineer.com/question.md?key=semester_8%2Fmultimedia_system%3Aimage-recognition-steps) — Multimedia System, chapter 3, asked 2× - [JPEG compression process steps](https://www.bctengineer.com/question.md?key=semester_8%2Fmultimedia_system%3Ajpeg-compression-steps) — Multimedia System, chapter 5, asked 3× ## Related in other subjects - [Nitrogen cooled: entropy generation](https://www.bctengineer.com/question.md?key=semester_2%2Ffundamental_of_thermodynamics_and_heat_transfer%3Anitrogen-cooling-entropy-generation) — Fundamental of Thermodynamics & Heat Transfer, chapter 5 - [Biogas generation stages and affecting factors](https://www.bctengineer.com/question.md?key=semester_6%2Fenergy_resources%3Abiogas-generation-stages-factors) — Energy Resources, chapter 3 - [Process of pipe laying and jointing](https://www.bctengineer.com/question.md?key=semester_5%2Fwater_supply_engineering%3Apipe-laying-jointing) — Water Supply Engineering, chapter 8, asked 7× - [Process control block](https://www.bctengineer.com/question.md?key=semester_6%2Foperating_system%3Apcb-explain) — Operating System, chapter 2