New plugin request - keep only the audio specified in the profile in radarr/sonarr

[Jorman]

Would it be possible, or rather, do you think it would be useful to have a plugin that analyzes the audio profile of Radarr and Sonarr, obviously depending on the library, and that keeps only the audio of the corresponding profile, plus any und?

I am currently doing something similar but with an external script, for some recompressions, basically I do it this way:

I analyze the download and retrieve the audio profile information, from Radarr or Sonarr, depending on the file to be managed.
Then I analyze the file and analyze each track to understand what language it is.
Then I remove everything that is not in the same language as the Radarr or Sonarr language profile, but I also keep the unds, precisely because I don't know what they are. If there are multiple audio tracks in the desired language, I keep them all.

Does that make sense? Is there a plugin that does this?

[Marnalas]

This looks interesting.

There are already ways to delete all audio streams, except for specific languages. Example: community plugin Tdarr_Plugin_MC93_Migz3CleanAudio

Using flow plugins, the only thing missing to do what you want is to retrieve the language tagged in the radarr/sonarr profile. I already have made a plugin that gets information from radarr/sonarr: #812
I would only need to add the logic to retrieve the language code from the profile.

I will look into it.

EDIT:
In Radarr I do see a language field in the profile but not in Sonarr. What do you use on the latter ?
If you could provide your script (at least the part that handles the retrieval), I could see what API endpoint you use.

[Jorman]

Hi, of course, thank you for considering this.
I have no problem sharing, in fact, it's a way to learn, especially for me since I'm not a professional programmer!

While I'm here, I'll ask you a question, I'm sure you can answer it:
Is there a plug-in for Tdarr to run an external script? Scrypt in Python or Docker or Bash, etc.

I'm asking you this because this plug-in that interfaces with Radarr/Sonarr that you're looking at and that I asked about has a limitation, but correct me if I'm wrong, namely:
If I'm not mistaken, Tdarr plug-ins "understand" the language based on tags, but if I'm not mistaken, language tags are only available for mkv and mp4 containers, other containers either have limitations or do not support language tags.
For this reason, in my script I use an external program to determine the language of the stream, so I can check also avi and any video extension. Just for your curiosity, I use this script that I made, and for now it does its job well -> https://github.com/Jorman/Scripts/tree/master/AudioMediaChecker
A little note, there's a flag to return a json format, for now is not under the documentation, I use the script in this way:

def _analyze_audio_languages_with_docker(src: Path) -> list[dict]:
    host_prefix, container_prefix = docker_mount_point.split(":", 1)
    p = src.resolve().as_posix()
    if p.startswith(host_prefix):
        container_file = container_prefix + p[len(host_prefix):]
    else:
        container_file = p  # if it is not in the mount, use as is (no assert)

    cmd = [
        "docker", "run", "--rm", "--gpus", "all",
        "-v", docker_mount_point,
        "-v", whisperai_model_mount_point,
        "chryses/audiomedia-checker:latest",
        "--gpu", "--check-all-tracks", "--dry-run", "--json",
        "--file", container_file,
    ]
    res = subprocess.run(cmd, capture_output=True, text=True, check=True)
    return json.loads(res.stdout.strip())  # es.: [{"track":1,"language":"ita"}, ...]

So, to extract data from Radarr/Sonarr, I use this. I admit that Sonarr is strange, and perhaps what I have done is not functional. It could certainly be improved, but it works, at least for now!

def get_profile_language_iso639_2(app, api_key, host, port, entity_id):
    base = host.rstrip("/")
    if port and f":{port}" not in base:
        base = f"{base}:{port}"
    headers = {"X-Api-Key": api_key}

    if app == "radarr":
        movie = requests.get(f"{base}/api/v3/movie/{entity_id}", headers=headers).json()
        qpid = movie["qualityProfileId"]
        qp = requests.get(f"{base}/api/v3/qualityprofile/{qpid}", headers=headers).json()
        lang_name = qp["language"]["name"]
        return pycountry.languages.lookup(lang_name).alpha_3.lower()

    elif app == "sonarr":
        # 1) take Quality Profile from the series
        series = requests.get(f"{base}/api/v3/series/{entity_id}", headers=headers).json()
        qpid = series["qualityProfileId"]
        qp = requests.get(f"{base}/api/v3/qualityprofile/{qpid}", headers=headers).json()

        # 2) Custom Formats map and find the LanguageSpecification CF with the highest score
        all_cf = requests.get(f"{base}/api/v3/customformat", headers=headers).json()
        cf_by_id = {cf["id"]: cf for cf in all_cf}

        best = None  # (score, lang_id)
        for it in qp.get("formatItems", []):
            ref = it.get("format")
            if isinstance(ref, int):
                cf_id = ref
            elif isinstance(ref, dict):
                cf_id = ref.get("id") or ref.get("customFormatId")
            else:
                continue
            cf = cf_by_id.get(cf_id)
            if not cf:
                continue
            for spec in cf.get("specifications", []):
                impl = (spec.get("implementation") or spec.get("implementationName") or "")
                if "LanguageSpecification" not in impl:
                    continue
                fields = spec.get("fields")
                val = None
                if isinstance(fields, dict):
                    val = fields.get("value")
                elif isinstance(fields, list):
                    for f in fields:
                        if isinstance(f, dict) and f.get("name") == "value":
                            val = f.get("value")
                            break
                if val is None:
                    continue
                score = it.get("score", 0)
                if best is None or score > best[0]:
                    best = (score, int(val))

        # 3) manage “Original Language” or absence
        if not best or best[1] < 0:
            return "und"

        lang_id = best[1]

        # 4) resolve ID -> canonical name via Sonarr API
        langs = requests.get(f"{base}/api/v3/language", headers=headers).json()
        lang_name = None
        for L in langs:
            if L.get("id") == lang_id:
                lang_name = L.get("name")
                break

        if not lang_name:
            return "und"

        # 5) normalizes a few known cases and reverts to ISO 639-2
        norm = {
            "flemish": "dutch",
            "portuguese (brazil)": "portuguese",
        }
        key = lang_name.lower()
        lang_for_lookup = norm.get(key, lang_name)

        return pycountry.languages.lookup(lang_for_lookup).alpha_3.lower()

    else:
        raise ValueError("app must be 'radarr' o 'sonarr'")

And the entity_id is:
for Sonarr -> sonarr_series_id
for Radarr -> radarr_movie_id

Let me know if I can be of any further assistance.

[Marnalas]

Thank you for providing details.
Your AudioMediaChecker script is an interesting use of AI!

Your approach to retrieving language information from Sonarr's quality profile is ingenious. However, I have some concerns about its generalizability. While it works well for your specific configuration of quality profiles and custom formats, Sonarr offers so many configuration options that this method likely wouldn't work reliably for most users. Tdarr plugins ideally need to work across the vast majority (if not all) of user configurations, which makes this approach challenging to implement as a general solution.

You're absolutely right about the limitations of tag-based audio stream handling in Tdarr plugins. Files with inaccurate or missing tags will inevitably cause issues with these plugins.
The Radarr/Sonarr community's established practice for handling such cases is to exclude problematic files upstream, rather than complicating the transcoding process. I realize this is easier said than done and isn't always feasible, but it remains the recommended approach.

Because of these two points, it makes sense to keep your scripts as is. Sadly, I'm not aware of the capability to execute external scripts in Tdarr. I should mention that I'm just a community member who has written some plugins. I'm not part of the development team.
For authoritative answers, I'd recommend asking on the Tdarr Discord server or subreddit. Visit https://home.tdarr.io/help/ for the links.

[Jorman]

You are absolutely right, I agree that my approach is "limited."
I am currently working on its implementation, to correctly manage:
Original language.
Multiple languages.
Various normalizations.

This should make it more universal and return the correct values. For Sonarr, a list of all the languages in the profile. I realize that it is possible for a profile to have multiple languages, so it makes sense to keep all the languages in the profile.
Would you like me to update you when I'm done?

However, I believe that a plug-in that does what we said at the beginning could be interesting, if you can retrieve the information from Sonarr.
Obviously, it will be a plug-in with "limitations," meaning it will only work on MKV or MP4 files, not AVI files, and users must be informed that if the language tags are incorrect, the wrong audio tracks will be removed.
To be honest, this "problem" is already present in the plug-in you mentioned -> Tdarr_Plugin_MC93_Migz3CleanAudio.
Am I wrong?

Yes, it's true that Sonarr and Radarr try to filter releases as much as possible, and in fact I agree with the method, but those who make the releases don't always do everything right, especially outside the US.

I will definitely contact the team to ask if and how an external script can be run, because it would be useful to have "certainty" about the language of the audio track, at least that's what I think.

Thank you for your help!

However, I don't understand if you think the plug-in is feasible?

[supersnellehenk]

Using runCli you can run basically anything.

If all you want is 'native audio', I've made a plugin in the past which checks the original language of a file in Radarr/Sonarr. I still want to create a flow block version.