Cutconcat fixed max duration (#1292)

* CutConcatenate: added fixed max_duration parameter

* pre-commit applied

lhotse/dataset/cut_transforms/concatenate.py

@@ -12,22 +12,34 @@ class CutConcatenate:
     adding some silence between them to avoid a large number of padding frames that waste the computation.
     """
 
-    def __init__(self, gap: Seconds = 1.0, duration_factor: float = 1.0) -> None:
+    def __init__(
+        self,
+        gap: Seconds = 1.0,
+        duration_factor: float = 1.0,
+        max_duration: Optional[Seconds] = None,
+    ) -> None:
         """
         CutConcatenate's constructor.
 
         :param gap: The duration of silence in seconds that is inserted between the cuts;
             it's goal is to let the model "know" that there are separate utterances in a single example.
         :param duration_factor: Determines the maximum duration of the concatenated cuts;
             by default it's 1, setting the limit at the duration of the longest cut in the batch.
+        :param max_duration: If a value is given (in seconds), the maximum duration of concatenated cuts
+            is fixed to the value while duration_factor is ignored.
         """
         self.gap = gap
         self.duration_factor = duration_factor
+        self.max_duration = max_duration
 
     def __call__(self, cuts: CutSet) -> CutSet:
         cuts = cuts.sort_by_duration(ascending=False)
         return concat_cuts(
-            cuts, gap=self.gap, max_duration=cuts[0].duration * self.duration_factor
+            cuts,
+            gap=self.gap,
+            max_duration=self.max_duration
+            if self.max_duration
+            else cuts[0].duration * self.duration_factor,
         )