switchboard preparation not working

[RuABraun]

Having problems preparing switchboard, hoping someone here can help. At my institute we have these dirs for Switchboard:

Switchboard_2_Phase_II/     Switchboard_2_Phase_III/    Switchboard_Cellular_Part1/ Switchboard_Cellular_Part2

I get this error when running lhotse prepare with Switchboard_Cellular_Part1/

  File "/idiap/temp/rbraun/programs/anaconda3/lib/python3.7/site-packages/lhotse/recipes/switchboard.py", line 57, in prepare_switchboard                                                         
    groups.append({'audio': ap, 'text-0': name_to_text[f'{name}A'], 'text-1': name_to_text[f'{name}B']})                                                                                                   
KeyError: 'sw_40000A'

I notice that in Switchboard_Cellular_Part1/dbase/data I have an sph file sw_40000.sph, but I guess because it has no A the key error is occuring?

It's not clear to me whether I have weird versions of the corpus or if lhotse is doing something wrong.

[pzelasko]

I think you may have a different switchboard distribution that is not the target of the current recipe. The current recipe is just for phase I, please see description:

lhotse/lhotse/recipes/switchboard.py

Lines 4 to 9 in 115b564

	This is conversational telephone speech collected as 2-channel, 8kHz-sampled
	data. We are using just the Switchboard-1 Phase 1 training data.
	The catalog number LDC97S62 (Switchboard-1 Release 2) corresponds, we believe,
	to what we have. We also use the Mississippi State transcriptions, which
	we download separately from
	http://www.isip.piconepress.com/projects/switchboard/releases/switchboard_word_alignments.tar.gz

I think if you'd like to build and contribute a recipe for the other packages, you might find the existing recipe helpful as a starting point.

[RuABraun]

Yeah I saw the help, I thought Switchboard_Cellular_Part1/2 might correspond to Phase 1 but I guess not.

I will try and contribute a solution! Next week tho

edit: turns out other corpuses dont have transcripts, and I got access to the right corpus!