The reddots data collection for speaker recognition

A conventional speaker verification (SV) system needs an enrolment session to collect the training data. Li et al. (1997) introduced a speaker authentication method called verbal information verification (VIV) which verifies a speaker by verbal contents instead of speech characteristics. Such a system does not need an enrolment session. In this paper, VIV is combined with SV. We propose a system which uses VIV to collect training data during the first few accesses automatically, which are often from different acoustic environments. Then, a speaker dependent model is trained and speaker authentication can be performed by SV. This approach not only avoid formal enrolment session which brings convenience to the user, but mitigates the mismatch problem causing by different acoustic environments between training and test sessions. Our experiments show that the proposed system improved the SV performance over 40% compared to the conventional SV system.

Traditional speaker authentication focuses on speaker verification (SV) and speaker identification, which is accomplished by matching the speaker's voice with his or her registered speech patterns. In this paper, we propose a new technique, verbal information verification (VIV), in which spoken utterances of a claimed speaker are verified against the key (usually confidential) information in the speaker's registered profile automatically; to decide whether the claimed identity should be accepted or rejected. Using the proposed sequential procedure involving three question-response turns, we achieved an error-free result in a telephone speaker authentication experiment with 100 speakers. We further propose a speaker authentication system by combining VIV with SV. In the system, a user is verified by VIV in the first four to five accesses, usually from different acoustic environments. During these uses, one of the key questions pertains to a pass-phrase for SV. The VIV system collects and verifies the pass-phrase utterance for use as training data for speaker model construction. After a speaker-dependent model is constructed, the system then migrates to SV. This approach avoids the inconvenience of a formal enrollment procedure, ensures the quality of the training data for SV, and mitigates the mismatch caused by different acoustic environments between training and testing. Experiments showed that the proposed system improved the SV performance by over 40% in equal-error rate compared to a conventional SV system.

Acoustical communication is one of the fundamental prerequisites for the existence of human society. Textual language has become extremely important in modern life, but speech has dimensions of richness that text cannot approximate. From speech alone, fairly accurate guesses can be made as to whether the speaker is male or female, adult or child. In addition, experts can extract from speech information regarding e.g. the speaker’s state of mind. As computer power increased and knowledge about speech signals improved, research of speech processing became aimed at automated systems for many purposes. Speaker recognition is the complement of speech recognition. Both techniques use similar methods of speech signal processing. In automatic speech recognition, the speech processing approach tries to extract linguistic information from the speech signal to the exclusion of personal information. Conversely, speaker recognition is focused on the characteristics unique to the individual, disregarding the current word spoken. The uniqueness of an individual’s voice is a consequence of both the physical features of the person vocal tract and the person mental ability to control the muscles in the vocal tract. An ideal speaker recognition system would use only physical features to characterize speakers, since these features cannot be easily changed. However, it is obvious that the physical features as vocal tract dimensions of an unknown speaker cannot be simply measured. Thus, numerical values for physical features or parameters would have to be derived from digital signal processing parameters extracted from the speech signal. Suppose that vocal tracts could be effectively represented by 10 independent physical features, with each feature taking on one of 10 discrete values. In this case, 1010 individuals in the population (i.e., 10 billion) could be distinguished whereas today’s world population amounts to approximately 7 billion individuals. People can reliably identify familiar voices. About 2-3 seconds of speech is sufficient to identify a voice, although performance decreases for unfamiliar voices. One review of human speaker recognition (Lancker et al., 1985) notes that many studies of 8-10 speakers (work colleagues) yield in excess of 97% accuracy if a sentence or more of the test speech is heard. Performance falls to about 54% when duration is shorter than 1 second and/or distorted e.g., severely highpass or lowpass filtered. Performance also falls significantly if training and test utterances are processed through different transmission systems. A study

A text-dependent speaker verification system has been implemented for forensic application since 2008 when speech recording is allowed as legal evidence in Indonesian court. Due to the laborious procedures of the current system, an automatic text-independent speaker verification system is being developed in order to accelerate the forensic speaker verification process. The new forensic system is developed using a database spoken in Indonesian. The database is built on two scenarios, interview and conversation. This is done to simulate the real forensic condition for speaker verification in Indonesia. The system adopts the very famous Gaussian Mixture Model (GMM) and Universal Background Model (UBM) framework in speaker verification field and using Mel-Frequency Cepstral Coefficients (MFCC) as the speech feature. Along with the application of zero-normalization, the new forensic speaker verification system has successfully reached an Equal Error Rate (EER) of 4.66 % that is a better performance than the previous developed system of Indonesian forensic speaker verification.

A new large margin nearest neighborhood metric learning (LMNN) method for i-vector based speaker verification is proposed in this paper. In general, a verification decision is taken by computing the cosine distance between the i-vectors of the test utterance and the claimed identity. LMNN metric is learned from the examples and can be viewed as a linear transformation of the input i-vector space of the training and test utterance. In this work, the metric is learned with the objective of reducing the distance between the i-vectors of same class of speaker, while impostors are separated by a large margin. The metric learned in this manner leads to a better speaker verification performance. Speaker verification experiments are then conducted on the NIST 2008 and YOHO speaker verification databases. Experimental results indicate a reasonable improvement in performance, when compared to i-vector based speaker verification methods which use conventional cosine scoring.

Discriminative subspace modeling of SNR and duration variabilities for robust speaker verification

LSTM-based speaker verification usually uses a fixed-length local segment randomly truncated from an utterance to learn the utterance-level speaker embedding, while using the average embedding of all segments of a test utterance to verify the speaker, which results in a critical mismatch between testing and training. This mismatch degrades the performance of speaker verification, especially when the durations of training and testing utterances are very different. To alleviate this issue, we propose the deep segment attentive embedding method to learn the unified speaker embeddings for utterances of variable duration. Each utterance is segmented by a sliding window and LSTM is used to extract the embedding of each segment. Instead of only using one local segment, we use the whole utterance to learn the utterance-level embedding by applying an attentive pooling to the embeddings of all segments. Moreover, the similarity loss of segment-level embeddings is introduced to guide the segment attention to focus on the segments with more speaker discriminations, and jointly optimized with the similarity loss of utterance-level embeddings. Systematic experiments on Tongdun and VoxCeleb show that the proposed method significantly improves robustness of duration variant and achieves the relative Equal Error Rate reduction of 50% and 11.54% , respectively.

The INTERSPEECH 2020 Far-Field Speaker Verification Challenge (FFSVC 2020) addresses three different research problems under well-defined conditions: far-field text-dependent speaker verification from single microphone array, far-field text-independent speaker verification from single microphone array, and far-field text-dependent speaker verification from distributed microphone arrays. All three tasks pose a cross-channel challenge to the participants. To simulate the real-life scenario, the enrollment utterances are recorded from close-talk cellphone, while the test utterances are recorded from the far-field microphone arrays. In this paper, we describe the database, the challenge, and the baseline system, which is based on a ResNet-based deep speaker network with cosine similarity scoring. For a given utterance, the speaker embeddings of different channels are equally averaged as the final embedding. The baseline system achieves minDCFs of 0.62, 0.66, and 0.64 and EERs of 6.27%, 6.55%, and 7.18% for task 1, task 2, and task 3, respectively.

Besides background noise, channel effect and speaker’s health condition, emotion is another factor which may influence the performance of a speaker verification system. In this paper, the performance of a GMM-UBM based speaker verification system on emotional speech is studied. It is found that speech with various emotions aggravates the verification performance. Two reasons for the performance aggravation are analyzed, they are mismatched emotions between the speaker models and the test utterances, and the articulating styles of certain emotions which create intense intra-speaker vocal variability. In response to the first reason, an emotion-dependent score normalization method is proposed, which is borrowed from the idea of Hnorm. Index Terms: speaker verification, emotional speech

The automatic identification of person's identity from their voice is a part of modern telecommunication services. In order to execute the identification task, speech signal has to be transmitted to a remote server. So a performance of the recognition/identification system can be influenced by various distortions that occur when transmitting speech signal through a communication channel. This paper studies an effect of telecommunication channel, particularly commonly used narrowband (NB) speech codecs in current telecommunication networks, on a performance of automatic speaker recognition in the context of a channel/codec mismatch between enrollment and test utterances. An influence of speech coding on speaker identification is assessed by using the reference GMM-UBM method. The results show that the partially mismatched scenario offers better results than the fully matched scenario when speaker recognition is done on speech utterances degraded by the different NB codecs. Moreover, deploying EVS and G.711 codecs in a training process of the recognition system provides the best success rate in the fully mismatched scenario. It should be noted here that the both EVS and G.711codecs offer the best speech quality among the codecs deployed in this study. This finding also fully corresponds with the finding presented by Janicki & Staroszczyk in [1] focusing on other speech codecs.

Voice activity detection plays an important role for an efficient voice interface between human and mobile devices, since it can be used as a trigger to activate an automatic speech recognition module of a mobile device. If the input speech signal can be recognized as a predefined magic word coming from a legitimate user, it can be utilized as a trigger. In this paper, we propose a voice trigger system using a keyword-dependent speaker recognition technique. The voice trigger must be able to perform keyword recognition, as well as speaker recognition, without using computationally demanding speech recognizers to properly trigger a mobile device with low computational power consumption. We propose a template based method and a hidden Markov model (HMM) based method for the voice trigger to solve this problem. The experiments using a Korean word corpus show that the template based method performed 4.1 times faster than the HMM based method. However, the HMM based method reduced the recognition error by 27.8% relatively compared to the template based method. The proposed methods are complementary and can be used selectively depending on the device of interest. <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup>

The results of a study aimed at finding the importance of pitch for automatic speaker recognition are presented. Pitch contours were obtained for 60 utterances, each approximately 2-sec in duration, of 10 female speakers. A data-reduction procedure based on the Karhunen-Loeve representation was found effective in representing the pitch information in each contour in a 20-dimensional space. The data were divided into two portions; one part was used to design the speaker recognition system, while the other part was used to test the effectiveness of the design. The 20-dimensional vectors representing the pitch contours of the design set were linearly transformed so that the ratio of interspeaker to intraspeaker variance in the transformed space was maximum. A reference utterance was formed for each speaker by averaging the transformed vectors of that speaker. The test utterance was assigned to the speaker corresponding to the reference utterance with the smallest Euclidean distance in the transformed space. The percentage of correct identifications using the above procedure was found to be 97%. The recognition rate increased to 98% with the addition of duration as an independent measure.

Probabilistic linear discriminant analysis (PLDA) is widely described as an effective model for text-independent speaker verification in the i-vector space. The PLDA scoring function is typically formulated as the likelihood ratio between the speaker-adapted and the universal PLDAs. In this case, the adaptation of PLDA was performed through the speaker factors. In this paper, we show that the channel factors of the PLDA could be equivalently exploited to deal with the multi-source conditions. In speaker verification, with the proposed method, a PLDAmodel trained on conversational telephone speech could be adequately adapted for interview-style microphone recordings. Experimental results on NIST SRE'08 and SRE'10 datasets confirm that the proposed method is effective, especially for the case whereby enrollment and test utterances were captured from different sources.

We propose a method to improve speaker verification performance when a test utterance is very short. In some situations with short test utterances, performance of ivector/probabilistic linear discriminant analysis systems degrades. The proposed method transforms short-utterance feature vectors to adequate vectors using a deep neural network, which compensate for short utterances. To reduce the dimensionality of the search space, we extract several principal components from the residual vectors between every long utterance i-vector in a development set and its truncated short utterance i-vector. Then an input i-vector of the network is transformed by linear combination of these directions. In this case, network outputs correspond to weights for linear combination of principal components. We use public speech databases to evaluate the method. The experimental results on short2-10sec condition (det6, male portion) of the NIST 2008 speaker recognition evaluation corpus show that the proposed method reduces the minimum detection cost relative to the baseline system, which uses linear discriminant analysis transformed i-vectors as features.

An experiment was performed to find out how well human listeners could determine whether or not two different utterances were spoken by the same speaker. The speech was coded in three ways: high-quality PCM (natural speech), linear prediction encoding (LPC), and ADPCM at 24 kbps. For this experiment all combinations of these three coding methods were used. A group of 16 reference speakers (customers—8 males, 8 females) along with 78 test speakers (16 reference and 62 imposters) were used in the experiment. Two test utterances were used: one for the male speakers and one for the female speakers. The 30 naive listeners who participated in the experiment were required to make same/different judgments. The results will be discussed in terms of the imposter acceptance error rates (miss rates) and customer rejection error rates (false-alarm rates) for the different test conditions.