Episode title detection failure when dashes (-) are used as separators.

[SnoutBaron]

For example, this is completely undetectable to Taiga:
HeavenlyDelusion-S01E01-Heaven and Hell [1151EF50]

This is detectable:
HeavenlyDelusion_S01E01_Heaven and Hell [1151EF50]

The issue is that a lot of torrents use (-) as separators, so for the moment I'm using PowerRenamer to regex dashes to underscores, unless there's something in Taiga that I'm missing.

[quanticism]

I've been using Taiga for years and its title detection is impressive. I suspect training a local LLM to parse titles may be more resilient to unhandled edge cases, but I'm sure this idea has already crossed the mind of Taiga's devs.

[erengy]

See #1215, #1071, #1069, #683 for similar issues.

Currently fixed in Anitomy's development branch:

>anitomy "HeavenlyDelusion-S01E01-Heaven and Hell [1151EF50]"
┌──────────────────────────────────┐
│ Element       │ Value            │
│──────────────────────────────────│
│ title         │ HeavenlyDelusion │
│ season        │ 01               │
│ episode       │ 01               │
│ episode_title │ Heaven and Hell  │
│ file_checksum │ 1151EF50         │
└──────────────────────────────────┘

I suspect training a local LLM to parse titles may be more resilient to unhandled edge cases, but I'm sure this idea has already crossed the mind of Taiga's devs.

I thought about making use of ML, but it comes with its own challenges and it's hard to predict if it'd perform significantly better or not. Besides, you might still need a rule-based system for creating the initial training dataset. So I ended up rewriting Anitomy instead, but I'd be interested in the results if anyone else wants to go that route.