Language model adaptation using result selection

摘要: A received utterance is recognized using different language models. For example, recognition of the independently performed a baseline model (BLM) and an adapted (ALM). determination made as to what results from are more likely be accurate. Different features may used assist in making (e.g. scores, confidences, acoustic quality measurements, . ) used. classifier trained then determining whether select BLM or ALM. automatically re-trained that adjusts weight training data response differences between two obtained applying