Telecommunications & Signal Processing Laboratory
Multimedia Signal Processing
Speech Database
Version 2 (2018-11)
The TSP speech contains over 1400 utterances spoken by 24 speakers (half
male, half female). This report describes the TSP speech database. The data was
recorded in an anechoic room. The database
includes the original samples (48 kHz sampling rate), and also the data filtered
and subsampled to different sample rates. The data is covered by a permissive
Simplified BSD licence.
Documentation
48k.zip (248 MB): Speech files (1444) sampled at 48 kHz-
16k-LP7.zip 86 MB): Speech files (1444) sampled at 16 kHz, with a 7 kHz lowpass filter
-
8k-mIRSS.zip (42 MB): Speech files (1444) sampled at 8 kHz, with modified IRS (Send) filtering
-
8k-G712.zip (43 MB): Speech files (1444) sampled at 8 kHz, with G.712 filtering
-
Filters.zip: Filter coefficient files used to create the downsampled speech files
-
Scripts.zip: Scripts that were used to generate the downsampled speech files
Network Delay Traces
18 round-trip network delay traces were collected. 8 were from a PC in the
Telecommunications & Signal Processing (TSP) Laboratory at McGill University
communicating with PC's at other universities within Canada and 10 were from a
home PC in Montreal (connected by cable modem to the Internet) to other PC's
within Canada (also connected by cable modem).
A modified version of ping was used to continuously send 45-byte ICMP packets
every 10 ms. Each trace was for 1,000,000 packets, lasting a total of 2 hours 47
minutes. The traces contain the round-trip time recorded at the sender. A value
of 0 ms in the trace that the packet was lost and never returned to the sender.