A Robust Algorithm for Pitch Tracking ( RAPT )

By David Talkin. 2005

989 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/f3f1d8960a6fde5bc4dc15acbfdce21cfc9b7452

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Reformulating the HMM as a trajectory model by imposing explicit relationships between static and dynamic feature vector sequences (2004, methodology): “Fundamental frequency (F0) values were extracted by the ESPS get f0 (Talkin, 1995) for 5-ms intervals.”
Robust Speaker-Adaptive HMM-Based Text-to-Speech Synthesis (2009, methodology): “Third, values are re-extracted in the speaker-specific range using three methods: IFAS, a fixed-point analysis called TEMPO [46] and the ESPS get- tool [47], [48].”
Speech Enhancement for a Noise-Robust Text-to-Speech Synthesis System Using Deep Recurrent Neural Networks (2016, methodology): “7488/ds/1356 From the clean and the noisy speech database we extracted using STRAIGHT [26] 60 Mel cepstral coefficients, 25 band aperiodicity components and using SPTK [27] we extracted fundamental frequency (F0) and voiced/unvoiced information with the RAPT F0 extraction method [28].”
Data augmentation for low resource languages (2014, unknown): “The input to MLP was a 504-dimensional stack of 4 past, 1 current and 4 future vectors, where each vector was a 13-dimensional PLP feature vector augmented with pitch [40], its delta (Δ), delta-delta (Δ(2)) and triples (Δ(3)).”
Estimation of Glottal Closure Instants in Voiced Speech Using the DYPSA Algorithm (2007, background): “characteristics of voiced speech including spectral quasi-stationarity and the periodic behavior of the vocal folds [38].”
Automatic speech emotion recognition using modulation spectral features (2011, methodology): “Pitch is computed for voiced speech using the pitch tracking algorithm in (Talkin, 1995), and intensity is measured for”
Investigating RNN-based speech enhancement methods for noise-robust Text-to-Speech (2016, unknown): “Using STRAIGHT [27] we extracted 60 MCEP coefficients, 25 BAP components and using SPTK [28] we extracted F0 and V/UV information with the RAPT F0 extraction method [29].”
Neural Network Bottleneck Features for Language Identification (2014, methodology): “In addition to the 476 FDLP features, a pitch value is estimated for each frame us-ing the RAPT algorithm [13] followed by a speaker-based mean and variance normalization.”
PYIN: A fundamental frequency estimator using probabilistic threshold distributions (2014, background): “Despite the high success rate of existing algorithms, this procedure is generally flawed because (potentially correct) frequency candidates that were discarded in the frame-wise stage cannot be recovered in the smoothing stage, and hence multiple frame-wise pitch candidates should be used before smo”
Speaker-Dependent WaveNet Vocoder (2017, methodology): “The fo sequences were extracted by RAPT algorithm [21].”
Crepe: A Convolutional Representation for Pitch Estimation (2018, methodology): “Those functions include the cepstrum [6], the autocorrelation function (ACF) [7], the average magnitude difference function (AMDF) [8], the normalized cross-correlation function (NCCF) as proposed by RAPT [9] and PRAAT [10], and the cumulative mean normalized difference function as proposed by YIN [”
Speaker-independent raw waveform model for glottal excitation (2018, methodology): “F0 was estimated from the synthetic signals using the RAPT algorithm [26], and log-domain F0 difference (in cents: 100 cents is one semitone, 12 semitones is one octave) is reported over frames where the voicing estimates agree.”
Speech Enhancement of Noisy and Reverberant Speech for Text-to-Speech (2018, methodology): “(F0) and voiced/unvoiced (V/UV) information using SPTK [39] F0 extraction routine with the RAPT’s extraction method option [40].”
Text-to-Speech Synthesis (2009, unknown): “trackers include Talkin’s get f0 and RAPT algorithms [431] (these were not included in Bags h w’s test, but are generally acknowledged to be very accurate).”
Analog acoustic expression in speech communication (2006, methodology): “F0 measurements were extracted using an autocorrelation pitch extraction algorithm (Talkin, 1995) with a 75 ms autocorrelation window.”
Bunched LPCNet : Vocoder for Low-cost Neural Text-To-Speech Systems (2020, methodology): “(2) We used the RAPT algorithm [17] for pitch tracking.”
Semi-supervised training of Deep Neural Networks (2013, methodology): “The acoustic models (both GMM-HMM and DNNs) are trained on VTLN-warped 12-dimensional PLPs augmented by C0, pitch (F0) [19], and probability of voicing with logit transformation.”