The Journal of the Acoustical Society of America · 1971 · 683 citations · 0 references
EngineeringSpeaker NormalizationPhonologySpeech RecognitionPhoneticsNormalization QualityRussian Vowels SpokenRussian VowelsLanguage StudiesMorphologySpeech CommunicationSpeech TechnologyText NormalizationPhonology MorphologyLanguage RecognitionSpeech ProcessingSpeech PerceptionLinguisticsSpeaker Recognition
Existing formant normalization methods make F1N and F2N invariant to compression, expansion, and parallel shift. The study investigates speaker normalization for Russian vowel classification and proposes a new statistical normalization method. The authors compare existing and the proposed normalization methods using a quality index η and demonstrate vowel classification via a self‑normalized formant plot. The proposed normalization achieves the highest quality index overall and for each vowel pair, enabling effective Russian vowel classification.
The problem of speaker normalization is investigated for classifying the Russian vowels. The known methods of normalization of formants F1 and F2 which make their normalized values F1N and F2N invariant to compression or expansion and to parallel shift are mentioned. A new statistical method of normalization is then suggested. All of these methods are compared using an index of normalization quality (η). The result of the comparison shows that the normalization suggested in this letter has the largest index, not only on the average but also for each individual vowel pair. The procedure of classification of the Russian vowels using a self-normalized formant plot is shown at the end of the letter.