Hypothesis Clustering and Merging: Novel MultiTalker Speech Recognition with Speaker Tokens