Why local recording changes the result
A normal remote interview is recorded from the call itself, which means every voice has already been squeezed through a conferencing codec and battered by whatever the connection was doing at the time. Dropouts, robotic patches and swallowed syllables are baked in before transcription starts.
Riverside sidesteps that by recording each person on their own machine and uploading the files afterwards. What you end up with is the audio as their microphone heard it, unaffected by the network, which removes the single most common cause of poor remote transcripts.
Transcribe the tracks separately
Because each participant arrives as a separate file, you can transcribe them one at a time and know with certainty who said every line. That is more reliable than any automatic speaker labelling, and it means crosstalk stops mattering: each person's track contains only their own voice.
Merge the results using the timestamps that come with every transcript. For an interview or a two-person podcast this is quick work, and the finished document is accurate enough to quote from after a much lighter review pass than mixed audio would need.
Frequently asked questions
Should I upload the separate tracks or the combined file?
Separate tracks, whenever you have them. They remove crosstalk and tell you exactly who is speaking, which is worth the small extra effort of merging the transcripts afterwards using the timestamps.
Riverside already offers transcription. Why use this?
Mostly cost and control. This gives you a transcript with no subscription tied to the recording platform, exported as TXT, SRT or VTT files you keep, on audio processed on hardware we own.
The tracks are large. Will they upload?
Local recordings are high quality and therefore big. Pro handles two gigabytes and five hours per file, and uploading one track at a time keeps each upload well within that.