Defining training, development and test set partitions for speech corpora is usually accomplished by hand. However, for the dataset under investigation, which contains a large number of speakers, eight different languages and code-switching between all the languages, this style of partitioning is not feasible. Therefore, we view the partitioning task as a resource allocation problem and propose to solve it automatically and optimally by the application of mixed-integer linear programming. Using this approach, we are able to partition a new 41.6-hour multilingual corpus of code-switched speech into training, development and testing partitions while maintaining a fixed number of speakers and a specific amount of code-switched speech in the development and test partitions. For this newly partitioned corpus, we present baseline speech recognition results using a state-of-the-art multilingual transformer model (Wav2Vec2-XLS-R) and show that the exclusion of very short utterances (<1s) results in substantially improved speech recognition performance.

This paper reports on the semi-supervised development of acoustic and language models for under-resourced, code-switched speech in five South African languages. Two approaches are considered. The first constructs four separate bilingual automatic speech recognisers (ASRs) corresponding to four different language pairs between which speakers switch frequently. The second uses a single, unified, five-lingual ASR system that represents all the languages (English, isiZulu, isiXhosa, Setswana and Sesotho). We evaluate the effectiveness of these two approaches when used to add additional data to our extremely sparse training sets. Results indicate that batch-wise semi-supervised training yields better results than a non-batch-wise approach. Furthermore, while the separate bilingual systems achieved better recognition performance than the unified system, they benefited more from pseudolabels generated by the five-lingual system than from those generated by the bilingual systems.

This paper considers the impact of automatic segmentation on the fully-automatic, semi-supervised training of automatic speech recog-nition (ASR) systems for five-lingual code-switched (CS) speech. Four automatic segmentation techniques were evaluated in terms ofthe recognition performance of an ASR system trained on the resulting segments in a semi-supervised manner. For comparative purposesa semi-supervised syste Three of these use a newly proposed convolutional neural network (CNN) model for framewise classification,and include a novel form of HMM smoothing of the CNN outputs. Automatic segmentation was applied in combination with automaticspeaker diarization. The best-performing segmentation technique was also evaluated without speaker diarization. An evaluation basedon 248 unsegmented soap opera episodes indicated that voice activity detection (VAD) based on a CNN followed by Gaussian mixturemodel-hidden Markov model smoothing (CNN-GMM-HMM) yields the best ASR performance. The semi-supervised system trainedwith the best automatic segmentation achieved an overall WER improvement of 1.1% absolute over a semi-supervised system trainedwith manually created segments. Furthermore, we found that recognition rates improved even further when the automatic segmentationwas used in conjunction with speaker diarization.

We present an analysis of semi-supervised acoustic and language model training for English-isiZulu code-switched (CS) ASR using soap opera speech. Approximately 11 hours of untranscribed multilingual speech was transcribed automatically using four bilingual CS transcription systems operating in English-isiZulu, English-isiXhosa, English-Setswana and English-Sesotho. These transcriptions were incorporated into the acoustic and language model training sets. Results showed that the TDNN-F acoustic models benefit from the additional semi-supervised data and that even better performance could be achieved by including additional CNN layers. Using these CNN-TDNN-F acoustic models, a first iteration of semi-supervised training achieved an absolute mixed-language WER reduction of 3.44%, and a further 2.18% after a second iteration. Although the languages in the untranscribed data were unknown, the best results were obtained when all automatically transcribed data was used for training and not just the utterances classified as English-isiZulu. Despite perplexity improvements, the semi-supervised language model was not able to improve the ASR performance.