Automatic Speech Recognition (CS753) Lecture 20: Pronunciation Modeling Instructor: Preethi Jyothi Oct 16, 2017
Pronunciation Dictionary/Lexicon Pronunciation model/dictionary/lexicon: Lists one or more pronunciations for a word Typically derived from language experts: Sequence of phones written down for each word Dictionary construction involves: 1. Selecting what words to include in the dictionary 2. Pronunciation of each word (also, check for multiple pronunciations)
Grapheme-based models
Graphemes vs. Phonemes Instead of a pronunciation dictionary, one could represent a pronunciation as a sequence of graphemes (or letters). That is, model at the grapheme level. Useful technique for low-resourced/under-resourced languages Main advantages: 1. Avoid the need for phone-based pronunciations 2. Avoid the need for a phone alphabet 3. Works pretty well for languages with a direct link between graphemes (letters) and phonemes (sounds)
Grapheme-based ASR Language ID System WER (%) Vit CN CNC Kurmanji Phonetic 67.6 65.8 205 Kurdish Graphemic 67.0 65.3 Tok Pisin 207 Cebuano 301 Kazakh 302 Telugu 303 Lithuanian 304 Phonetic 41.8 40.6 Graphemic 42.1 41.1 Phonetic 55.5 54.0 Graphemic 55.5 54.2 Phonetic 54.9 53.5 Graphemic 54.0 52.7 Phonetic 70.6 69.1 Graphemic 70.9 69.5 Phonetic 51.5 50.2 Graphemic 50.9 49.5 64.1 39.4 52.6 51.5 67.5 48.3 Image from: Gales et al., Unicode-based graphemic systems for limited resource languages, ICASSP 15
Graphemes vs. Phonemes Instead of a pronunciation dictionary, one could represent a pronunciation as a sequence of graphemes (or letters) Useful technique for low-resourced/under-resourced languages Main advantages: 1. Avoid the need for phone-based pronunciations 2. Avoid the need for a phone alphabet 3. Works pretty well for languages with a direct link between graphemes (letters) and phonemes (sounds)
Grapheme to phoneme (G2P) conversion
Grapheme to phoneme (G2P) conversion Produce a pronunciation (phoneme sequence) given a written word (grapheme sequence) Learn G2P mappings from a pronunciation dictionary Useful for: ASR systems in languages with no pre-built lexicons Speech synthesis systems Deriving pronunciations for out-of-vocabulary (OOV) words
G2P conversion (I) One popular paradigm: Joint sequence models [BN12] Grapheme and phoneme sequences are first aligned using EM-based algorithm Results in a sequence of graphones (joint G-P tokens) Ngram models trained on these graphone sequences WFST-based implementation of such a joint graphone model [Phonetisaurus] [BN12]:Bisani & Ney, Joint sequence models for grapheme-to-phoneme conversion,specom 2012 [Phonetisaurus] J. Novak, Phonetisaurus Toolkit
G2P conversion (II) Neural network based methods are the new state-of-the-art for G2P Bidirectional LSTM-based networks using a CTC output layer [Rao15]. Comparable to Ngram models. Incorporate alignment information [Yao15]. Beats Ngram models. No alignment. Encoder-decoder with attention. Beats the above systems [Toshniwal16].
LSTM + CTC for G2P conversion [Rao15] Model Word Error Rate (%) Galescu and Allen [4] 28.5 Chen [7] 24.7 Bisani and Ney [2] 24.5 Novak et al. [6] 24.4 Wu et al. [12] 23.4 5-gram FST 27.2 8-gram FST 26.5 Unidirectional LSTM with Full-delay 30.1 DBLSTM-CTC 128 Units 27.9 DBLSTM-CTC 512 Units 25.8 DBLSTM-CTC 512 + 5-gram FST 21.3 [Rao15] Grapheme-to-phoneme conversion using LSTM RNNs, ICASSP 2015
G2P conversion (II) Neural network based methods are the new state-of-the-art for G2P Bidirectional LSTM-based networks using a CTC output layer [Rao15]. Comparable to Ngram models. Incorporate alignment information [Yao15]. Beats Ngram models. No alignment. Encoder-decoder with attention. Beats the above systems [Toshniwal16].
Seq2seq models (with alignment information [Yao15]) Method PER (%) WER (%) encoder-decoder LSTM 7.53 29.21 encoder-decoder LSTM (2 layers) 7.63 28.61 uni-directional LSTM 8.22 32.64 uni-directional LSTM (window size 6) 6.58 28.56 bi-directional LSTM 5.98 25.72 bi-directional LSTM (2 layers) 5.84 25.02 bi-directional LSTM (3 layers) 5.45 23.55 Data Method PER (%) WER (%) CMUDict past results [20] 5.88 24.53 bi-directional LSTM 5.45 23.55 NetTalk past results [20] 8.26 33.67 bi-directional LSTM 7.38 30.77 Pronlex past results [20, 21] 6.78 27.33 bi-directional LSTM 6.51 26.69 [Yao15] Sequence-to-sequence neural net models for G2P conversion, Interspeech 2015
G2P conversion (II) Neural network based methods are the new state-of-the-art for G2P Bidirectional LSTM-based networks using a CTC output layer [Rao15]. Comparable to Ngram models. Incorporate alignment information [Yao15]. Beats Ngram models. No alignment. Encoder-decoder with attention. Beats the above systems [Toshniwal16]. [Rao15] Grapheme-to-phoneme conversion using LSTM RNNs, ICASSP 2015 [Yao15] Sequence-to-sequence neural net models for G2P conversion, Interspeech 2015 [Toshniwal16] Jointly learning to align and convert graphemes to phonemes with neural attention models, SLT 2016.
Encoder-decoder + attention for G2P [Toshniwal16] Attention Layer y t c t t Encoder Decoder h Tg h 3 h 2 h 1 d t x Tg x 3 x 2 x 1 [Toshniwal16] Jointly learning to align and convert graphemes to phonemes with neural attention models, SLT 2016.
Encoder-decoder + attention for G2P [Toshniwal16] Attention Layer c t t y t Data Method PER (%) CMUDict BiDir LSTM + Alignment [6] 5.45 DBLSTM-CTC [5] - DBLSTM-CTC + 5-gram model [5] - Encoder-decoder + global attn 5.04 ± 0.03 Encoder-decoder + local-m attn 5.11 ± 0.03 Encoder-decoder + local-p attn 5.39 ± 0.04 Ensemble of 5 [Encoder-decoder + global attn] models 4.69 Pronlex BiDir LSTM + Alignment [6] 6.51 Encoder-decoder + global attn 6.24 ± 0.1 Encoder-decoder + local-m attn 5.99 ± 0.11 Encoder-decoder + local-p attn 6.49 ± 0.06 NetTalk BiDir LSTM + Alignment [6] 7.38 Encoder-decoder + global attn 7.14 ± 0.72 Encoder-decoder + local-m attn 7.13 ± 0.11 Encoder-decoder + local-p attn 8.41 ± 0.19 Encoder Decoder h Tg h 3 h 2 h 1 d t x Tg x 3 x 2 x 1 [Toshniwal16] Jointly learning to align and convert graphemes to phonemes with neural attention models, SLT 2016.
Sub-phonetic feature-based models
Pronunciation Model Phone-Based Articulatory Features Each word is a sequence of phones Parallel streams of articulator movements Tends to be highly language dependent Based on theory of articulatory phonology 1 PHONE s eh n s 1 [C. P. Browman and L. Goldstein, Phonology 86]
Pronunciation Model Articulatory Features LIP- OPEN LIP- LOC TT-LOC TB-LOC TB- TT- OPEN OPEN VELUM Parallel streams of articulator movements Based on theory of GLOTTIS articulatory phonology 1 PHONE s eh n s LIP open/labial TON.TIP critical/alveolar mid/alveolar closed/alveolar critical/alveolar TON.BODY mid/uvular mid/palatal mid/uvular GLOTTIS open critical open VELUM closed open closed
Example: Pronunciations for word sense CANONICAL LIP TB TT GLOT VEL PHONE open/labial mid/uvular mid/palatal mid/uvular critical/alveolar mid/alveolar closed/alveolar critical/alveolar open critical open closed closed open s eh n s E.g. OBSERVED LIP TB TT GLOT VEL PHONE open/labial mid/uvular mid/palatal mid/uvular critical/alveolar mid/alveolar closed/alveolar critical/alveolar open critical open closed open closed s eh_n n t s Simple asynchrony across feature streams can appear as many phone alterations [Adapted from Livescu 05]
Dynamic Bayesian Networks (DBNs) Provides a natural framework to efficiently encode multiple streams of articulatory features Simple DBN with three random variables in each time frame A t-1 A t A t+1 B t-1 B t B t+1 C t-1 C t C t+1 frame t-1 frame t frame t+1
P. Jyothi, E. Fosler-Lussier & K. Livescu, Interspeech 12 DBN model of pronunciation Word Posn 0 1 2 3 Phn L-Lag s ao l v 0 s ao l v 1 - s ao l Trans Posn Phn Word solve s ao l v L-Lag T-Lag G-Lag L-Phn Prev- Phone T-Phn G-Phn Observed feature values Lip-Op TT-Op Glot sur Lip-Op sur TT-Op sur Glot
Factorized DBN model 1 Word Trans Posn Phn Set1 L-Lag T-Lag G-Lag Set2 L-Phn Prev- Phone T-Phn G-Phn Set3 Lip-Op TT-Op Glot Set4 sur Lip-Op sur TT-Op sur Glot Set5
Cascade of Finite State Machines Word Trans Posn L-Lag L-Phn sur Lipop sur TTop Prev- Phn Phn T-Lag T-Phn Lipop TTop G- Lag G- Phn Glot sur Glot Word Phn, Trans, L-Lag,T-Lag, G-Lag Lip-op, TT-op, Glot Posn L-Lag T-Lag G-Lag Prev-Phn Phn Phn, Trans L-Phn, T-Phn, G-Phn F1 F2 F3 F4 surlip-op, surtt-op, surglot F5 1 [P. Jyothi, E. Fosler-Lussier, K. Livescu, Interspeech 12]
Weighted Finite State Machine x1:y1/1.5 x2:y2/1.3 x4:y4/0.6 x3:y3/2.0
Weighted Finite State Machine x1:y1/1.5 x2:y2/1.3 x4:y4/0.6 x3:y3/2.0 w α (X, a) : weight of path a on input X. where α are learned parameters Linear model: w α (X, a) = α φ(x, a). where φ is a feature function. Decoding: For input X, find the path with minimum cost. a * = argmin w α (X, a) path a
Discriminative Training x1:y1/1.5 x2:y2/1.3 x4:y4/0.6 x3:y3/2.0 Online discriminative training algorithm to learn α Similar to structured perceptron [Collins 02]: Each training sample gives a decoded path and a correct path. Update α to bias towards correct path. Use a large-margin training algorithm adapted to work with a cascade of finite state machines 1 1 [P. Jyothi, E. Fosler-Lussier & K. Livescu, Interspeech-13]