Transcribe and stream podcast pages (any yt-dlp-extractable URL)

README.md

@@ -89,7 +89,7 @@ Your key is written to a git-ignored `.env` (never sent to the browser). Use `--
 | --- | --- |
 | `aai login` / `logout` / `whoami` | Manage the stored API key. |
 | `aai doctor` | Check your environment (API key, network, ffmpeg, microphone, agent tooling). |
-| `aai transcribe <file\|url>` | Transcribe a file, URL, or YouTube URL (`--sample`, `--llm`, `--show-code`). |
+| `aai transcribe <file\|url>` | Transcribe a file, URL, or YouTube/podcast page URL (`--sample`, `--llm`, `--show-code`). |
 | `aai transcripts list` / `get <id>` | Browse and fetch past transcripts. |
 | `aai stream [file]` | Real-time transcription from a file or the microphone. |
 | `aai agent` | *Run* a live two-way voice conversation (to **build** a voice agent app, use `aai init voice-agent`). |

aai_cli/code_gen/transcribe.py

@@ -40,28 +40,28 @@ def render(
     """
     if output is not None:
         llm_gateway = None  # `-o` returns before the chain runs in the real command
-    is_youtube = youtube.is_youtube_url(source)
+    needs_download = youtube.is_downloadable_url(source)
     parts = (
-        _header_block(llm_gateway, output, is_youtube=is_youtube)
-        + _transcribe_block(merged, source, is_youtube=is_youtube)
+        _header_block(llm_gateway, output, needs_download=needs_download)
+        + _transcribe_block(merged, source, needs_download=needs_download)
         + _result_block(merged, llm_gateway, output)
     )
     parts.append("")
     return "\n".join(parts)
 
 
 def _header_block(
-    llm_gateway: dict[str, object] | None, output: str | None, *, is_youtube: bool
+    llm_gateway: dict[str, object] | None, output: str | None, *, needs_download: bool
 ) -> list[str]:
     """Imports plus the api-key (and non-default environment) settings lines."""
     stdlib_imports = ["import os"]
-    if is_youtube:
-        # The YouTube path downloads audio to a temp dir before uploading.
+    if needs_download:
+        # The download path fetches audio to a temp dir before uploading.
         stdlib_imports += ["import tempfile"]
     if output == "json":
         stdlib_imports.insert(0, "import json")
     imports = ["import assemblyai as aai"]
-    if is_youtube:
+    if needs_download:
         imports.append("import yt_dlp")
     if llm_gateway:
         imports.append("from openai import OpenAI")
@@ -81,20 +81,20 @@ def _header_block(
     return parts
 
 
-def _transcribe_block(merged: dict[str, object], source: str, *, is_youtube: bool) -> list[str]:
+def _transcribe_block(merged: dict[str, object], source: str, *, needs_download: bool) -> list[str]:
     """The transcriber setup, optional config, the transcribe call, and error check."""
     parts = ["", "transcriber = aai.Transcriber()"]
     config_arg = ""
     if merged:
         kwargs = "\n".join(serialize.config_kwarg_lines(merged, indent=4))
         parts += ["", f"config = aai.TranscriptionConfig(\n{kwargs}\n)"]
         config_arg = ", config=config"
-    if is_youtube:
-        # AssemblyAI can't read a YouTube watch URL itself, so download the audio
-        # with yt-dlp into a temp dir and upload the local file — what the CLI does.
+    if needs_download:
+        # AssemblyAI can't read a YouTube/podcast page URL itself, so download the
+        # audio with yt-dlp into a temp dir and upload the local file — what the CLI does.
         parts += [
             "",
-            "# AssemblyAI can't fetch a YouTube URL itself; download the audio first.",
+            "# AssemblyAI can't fetch this page URL itself; download the audio first.",
             "with tempfile.TemporaryDirectory() as _tmp:",
             "    with yt_dlp.YoutubeDL(",
             '        {"format": "bestaudio/best", "outtmpl": f"{_tmp}/%(id)s.%(ext)s"}',

aai_cli/commands/stream.py

@@ -58,7 +58,7 @@ def _dispatch(session: StreamSession, opts: SourceOptions) -> None:
         # Raw PCM16 mono piped on stdin (e.g. `ffmpeg … -f s16le - | aai stream -`).
         stdin_src = StdinSource(sample_rate=opts.sample_rate or TARGET_RATE)
         session.run(stdin_src, stdin_src.sample_rate)
-    elif opts.source and youtube.is_youtube_url(opts.source):
+    elif opts.source and youtube.is_downloadable_url(opts.source):
         # Fetch the audio first, then stream the local file in real time.
         with tempfile.TemporaryDirectory(prefix="aai-yt-") as td:
             local = youtube.download_audio(opts.source, Path(td))
@@ -101,8 +101,8 @@ def stream(
     ctx: typer.Context,
     source: str | None = typer.Argument(
         None,
-        help="Audio file path, URL, or YouTube URL to stream. Use - for raw PCM16/mono/16k "
-        "on stdin. Omit to use the microphone.",
+        help="Audio file path, URL, or YouTube/podcast page URL to stream. Use - for raw "
+        "PCM16/mono/16k on stdin. Omit to use the microphone.",
     ),
     sample: bool = typer.Option(False, "--sample", help="Stream the hosted wildfires.mp3 sample."),
     # audio capture
@@ -395,9 +395,9 @@ def body(state: AppState, json_mode: bool) -> None:
             validate_sources(opts, has_llm=bool(llm_prompt), text_mode=text_mode)
             if opts.from_system_audio:
                 raise UsageError("--show-code does not support macOS system audio capture yet.")
-            if opts.source and youtube.is_youtube_url(opts.source):
+            if opts.source and youtube.is_downloadable_url(opts.source):
                 raise UsageError(
-                    "--show-code does not support YouTube sources yet.",
+                    "--show-code does not support downloaded sources (YouTube, podcast pages) yet.",
                     suggestion="Download the audio first (e.g. yt-dlp) and pass the local file.",
                 )
             code_source: str | None = None

aai_cli/commands/transcribe.py

@@ -66,6 +66,7 @@ def _validate_speakers_expected(merged: dict[str, object]) -> None:
             ("Transcribe a local file", "aai transcribe call.mp3"),
             ("Try it with the hosted sample", "aai transcribe --sample"),
             ("Transcribe a YouTube video", "aai transcribe https://youtu.be/dtp6b76pMak"),
+            ("Transcribe a podcast episode page", 'aai transcribe "https://podcasts.apple.com/…"'),
             ("Label who said what", "aai transcribe call.mp3 --speaker-labels"),
             ("Redact PII for compliance", "aai transcribe call.mp3 --redact-pii"),
             ("Summarize a recording", "aai transcribe call.mp3 --summarization"),
@@ -75,7 +76,7 @@ def _validate_speakers_expected(merged: dict[str, object]) -> None:
 )
 def transcribe(
     ctx: typer.Context,
-    source: str | None = typer.Argument(None, help="Audio file path, public URL, or YouTube URL."),
+    source: str | None = typer.Argument(None, help="Audio file, URL, or YouTube/podcast URL."),
     sample: bool = typer.Option(False, "--sample", help="Use the hosted wildfires.mp3 sample."),
     # model & language
     speech_model: aai.SpeechModel | None = typer.Option(
@@ -357,12 +358,12 @@ def transcribe(
         help="Print the equivalent Python SDK code and exit (does not transcribe).",
     ),
 ) -> None:
-    """Transcribe an audio file, URL, or YouTube link.
+    """Transcribe an audio file, URL, or YouTube/podcast link.
 
     Quickest start: aai transcribe call.mp3 (or --sample for the hosted demo).
 
-    Save with --out FILE, or pipe one field with -o text. A YouTube URL is downloaded
-    first, then transcribed.
+    Save with --out FILE, or pipe one field with -o text. YouTube and podcast-page
+    URLs (any page yt-dlp can extract) are downloaded first, then transcribed.
 
     Curated flags cover common features; --config KEY=VALUE and --config-file reach
     every other field. Analysis (summary, chapters, ...) renders in human mode.

aai_cli/onboard/sections.py

@@ -64,7 +64,7 @@ def first_request(prompter: Prompter, ctx: WizardContext) -> SectionResult:
     prompter.section("Your first transcription")
     api_key = config.resolve_api_key(profile=ctx.profile)
     source = prompter.text(
-        "Audio file path or YouTube URL (or press Enter to transcribe a sample clip)",
+        "Audio file path or YouTube/podcast URL (or press Enter to transcribe a sample clip)",
         default="",
     ).strip()
     label = source or "the sample clip"

aai_cli/skills/aai-cli/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: aai-cli
-description: Use the AssemblyAI CLI (`aai`) from the command line — transcribe audio/video files, URLs, and YouTube links; stream live real-time transcription from a mic/file/system audio; run full-duplex voice agents; query the LLM Gateway over transcripts; browse transcript and streaming-session history; sign in and manage account balance, usage, rate limits, API keys, and audit logs; scaffold a starter app (init); diagnose setup (doctor); and set up your coding agent's AssemblyAI docs MCP + skills (setup). Use whenever an agent is invoking the `aai` command.
+description: Use the AssemblyAI CLI (`aai`) from the command line — transcribe audio/video files, URLs, and YouTube/podcast links; stream live real-time transcription from a mic/file/system audio; run full-duplex voice agents; query the LLM Gateway over transcripts; browse transcript and streaming-session history; sign in and manage account balance, usage, rate limits, API keys, and audit logs; scaffold a starter app (init); diagnose setup (doctor); and set up your coding agent's AssemblyAI docs MCP + skills (setup). Use whenever an agent is invoking the `aai` command.
 ---
 
 # AssemblyAI CLI (`aai`)
@@ -72,8 +72,8 @@ agent," reach for `aai init voice-agent`, not `aai agent`.
 
 - **Build/scaffold an app (transcription, live captions, or a voice agent app)**
   → `aai init` — see `references/setup.md`
-- **Transcribe a file/URL/YouTube, stream live audio, run a live voice agent, or
-  query the LLM Gateway** → `references/transcription.md`
+- **Transcribe a file/URL/YouTube/podcast page, stream live audio, run a live
+  voice agent, or query the LLM Gateway** → `references/transcription.md`
 - **Browse past transcripts or streaming sessions** → `references/history.md`
 - **Sign in/out, identity, balance, usage, rate limits, API keys, audit log** →
   `references/account.md`

aai_cli/skills/aai-cli/references/transcription.md

@@ -4,9 +4,10 @@ Four commands. All accept `--json` (auto-enabled when piped) and `-o/--output`
 to print a single field. `transcribe`, `stream`, and `agent` accept
 `--show-code` to print equivalent Python SDK code without calling the API.
 
-## `aai transcribe [SOURCE]` — file / URL / YouTube
+## `aai transcribe [SOURCE]` — file / URL / YouTube / podcast page
 
-`SOURCE` is a local file path, public URL, or YouTube URL (downloaded first).
+`SOURCE` is a local file path, public URL, or a media-page URL yt-dlp can extract
+(YouTube, Apple Podcasts, Spreaker, SoundCloud, …) — those are downloaded first.
 Use `--sample` for the hosted `wildfires.mp3`. Analysis results (summary,
 chapters, sentiment, …) render automatically in human mode.
 
@@ -38,7 +39,7 @@ aai transcribe call.mp3 --show-code
 
 ## `aai stream [SOURCE]` — live real-time transcription
 
-Omit `SOURCE` to use the microphone; pass a file/URL/YouTube to stream that, or
+Omit `SOURCE` to use the microphone; pass a file/URL/media page to stream that, or
 `--sample`. macOS can capture system audio with `--system-audio` (mic + system)
 or `--system-audio-only`.
 

aai_cli/transcribe_exec.py

@@ -69,8 +69,8 @@ def run_transcription(
             return client.transcribe(api_key, str(local), config=transcription_config)
 
     audio = client.resolve_audio_source(source, sample=sample)
-    if youtube.is_youtube_url(audio):
-        # Fetch first; AssemblyAI can't read a YouTube watch URL itself.
+    if youtube.is_downloadable_url(audio):
+        # Fetch first; AssemblyAI can't read a YouTube/podcast page URL itself.
         with tempfile.TemporaryDirectory(prefix="aai-yt-") as td:
             local = youtube.download_audio(audio, Path(td))
             return client.transcribe(api_key, str(local), config=transcription_config)

aai_cli/youtube.py

@@ -1,3 +1,9 @@
+"""Downloading audio from media-page URLs (YouTube, podcast pages, …) via yt-dlp.
+
+The AssemblyAI API fetches direct audio URLs itself; this module handles the URLs it
+can't — HTML pages whose audio yt-dlp knows how to extract.
+"""
+
 from __future__ import annotations
 
 import logging
@@ -27,6 +33,34 @@ def is_youtube_url(source: str | None) -> bool:
     return bool(_YOUTUBE_RE.match(source.strip()))
 
 
+def is_downloadable_url(source: str | None) -> bool:
+    """True if `source` is a media-page URL whose audio must be downloaded first.
+
+    YouTube is matched by shape alone — no yt-dlp import needed, so a missing yt-dlp
+    still routes to ``download_audio``'s install hint. Other http(s) URLs match when
+    a dedicated yt-dlp extractor claims them (Apple Podcasts, Spreaker, SoundCloud,
+    …). Direct audio URLs and unknown pages match only yt-dlp's catch-all ``Generic``
+    extractor, which is excluded: those pass through untouched for the API to fetch.
+    """
+    if is_youtube_url(source):
+        return True
+    url = (source or "").strip()
+    if not url.startswith(("http://", "https://")):
+        # Local paths (and other non-URL sources) never need a download; skipping the
+        # extractor sweep also avoids importing yt-dlp on the common local-file path.
+        return False
+    return _ytdlp_extractor_claims(url)
+
+
+def _ytdlp_extractor_claims(url: str) -> bool:
+    """True if a dedicated (non-``Generic``) yt-dlp extractor matches `url`."""
+    try:
+        from yt_dlp.extractor import gen_extractor_classes
+    except ImportError:
+        return False
+    return any(ie.suitable(url) and ie.ie_key() != "Generic" for ie in gen_extractor_classes())
+
+
 def download_audio(url: str, dest_dir: Path) -> Path:
     """Download the best audio track of `url` into `dest_dir` and return its path.
 

tests/__snapshots__/test_cli_output_snapshots.ambr

@@ -704,9 +704,9 @@
    instead, pipe the text out with -o text | aai llm -f "…".
 
   ╭─ Arguments ──────────────────────────────────────────────────────────────────╮
-  │   source      [SOURCE]  Audio file path, URL, or YouTube URL to stream. Use  │
-  │                         - for raw PCM16/mono/16k on stdin. Omit to use the   │
-  │                         microphone.                                          │
+  │   source      [SOURCE]  Audio file path, URL, or YouTube/podcast page URL to │
+  │                         stream. Use - for raw PCM16/mono/16k on stdin. Omit  │
+  │                         to use the microphone.                               │
   ╰──────────────────────────────────────────────────────────────────────────────╯
   ╭─ Options ────────────────────────────────────────────────────────────────────╮
   │ --sample                          Stream the hosted wildfires.mp3 sample.    │
@@ -837,20 +837,19 @@
 
    Usage: aai transcribe [OPTIONS] [SOURCE]
 
-   Transcribe an audio file, URL, or YouTube link.
+   Transcribe an audio file, URL, or YouTube/podcast link.
 
    Quickest start: aai transcribe call.mp3 (or --sample for the hosted demo).
 
-   Save with --out FILE, or pipe one field with -o text. A YouTube URL is
-   downloaded
-   first, then transcribed.
+   Save with --out FILE, or pipe one field with -o text. YouTube and podcast-page
+   URLs (any page yt-dlp can extract) are downloaded first, then transcribed.
 
    Curated flags cover common features; --config KEY=VALUE and --config-file
    reach
    every other field. Analysis (summary, chapters, ...) renders in human mode.
 
   ╭─ Arguments ──────────────────────────────────────────────────────────────────╮
-  │   source      [SOURCE]  Audio file path, public URL, or YouTube URL.         │
+  │   source      [SOURCE]  Audio file, URL, or YouTube/podcast URL.             │
   ╰──────────────────────────────────────────────────────────────────────────────╯
   ╭─ Options ────────────────────────────────────────────────────────────────────╮
   │ --sample                                         Use the hosted              │
@@ -969,6 +968,8 @@
    $ aai transcribe --sample
    Transcribe a YouTube video
    $ aai transcribe https://youtu.be/dtp6b76pMak
+   Transcribe a podcast episode page
+   $ aai transcribe "https://podcasts.apple.com/…"
    Label who said what
    $ aai transcribe call.mp3 --speaker-labels
    Redact PII for compliance

tests/test_code_gen.py

@@ -118,6 +118,18 @@ def test_transcribe_render_youtube_downloads_before_upload():
     assert 'transcribe("https://www.youtube.com' not in code
 
 
+def test_transcribe_render_podcast_page_downloads_before_upload():
+    # Podcast episode pages (extractor-matched, like YouTube) generate the same
+    # download-first script; the page URL never reaches transcribe().
+    url = "https://podcasts.apple.com/us/podcast/some-show/id1535809341?i=1000123456789"
+    code = code_gen.transcribe({}, source=url)
+    ast.parse(code)
+    assert "import yt_dlp" in code
+    assert f"extract_info({url!r}, download=True)" in code
+    assert "transcriber.transcribe(_audio)" in code
+    assert "transcribe('https://podcasts.apple.com" not in code
+
+
 def test_transcribe_render_youtube_passes_config_to_local_upload():
     # With a config object the download still wraps the upload, and config flows through.
     code = code_gen.transcribe({"speaker_labels": True}, source="https://youtu.be/abc123")

tests/test_onboard_sections.py

@@ -34,6 +34,7 @@ def __init__(self, *, select: str = "skip", confirm: bool = True, text: str = "k
         self._confirm = confirm
         self._text = text
         self.confirm_defaults: list[bool] = []
+        self.text_titles: list[str] = []
 
     def section(self, title: str) -> None:
         pass
@@ -51,6 +52,7 @@ def select(
         return self._select
 
     def text(self, title: str, *, default: str | None = None) -> str:
+        self.text_titles.append(title)
         return self._text
 
 
@@ -118,10 +120,13 @@ def _fake(
     monkeypatch.setattr(transcribe_exec, "run_transcription", _fake)
     monkeypatch.setattr(transcribe_render, "render_transcript_result", lambda *a, **k: None)
     status_messages = _capture_status(monkeypatch)
-    assert sections.first_request(_ScriptedPrompter(text="meeting.mp3"), ctx) is SectionResult.DONE
+    prompter = _ScriptedPrompter(text="meeting.mp3")
+    assert sections.first_request(prompter, ctx) is SectionResult.DONE
     assert seen["source"] == "meeting.mp3"
     assert seen["sample"] is False
     assert status_messages == ["Transcribing meeting.mp3…"]
+    # The prompt advertises every accepted source kind, including podcast pages.
+    assert any("YouTube/podcast URL" in t for t in prompter.text_titles)
 
 
 def test_first_request_handles_failure(ctx: WizardContext, monkeypatch: pytest.MonkeyPatch) -> None: