fix(openai-chat): read ogg, opus, flac, aac, aiff, webm and mpeg input_audio as their true media types

request_from_openai_chat refused every input_audio format but wav and mp3,
before any provider was chosen, so a DSPy ogg/opus recording bound for Gemini
failed at ingest although Gemini takes audio/ogg natively. Each format now
reads as its true media type; an unknown format is still malformed, and the
Chat Completions and Anthropic builders still refuse audio at send (MAP-10).

The same change as the lm15-dev/lm15-python fix, on the line DSPy vendors;
CONTRACT_PIN stays at b721ce4, whose MAP-12 rule 4 still lists wav and mp3.

Signed-off-by: Jonas Dreyøe Herfort <jdreyoe@gmail.com>

Author: jonasherfort

docs/migrating-from-openai-chat.md

@@ -501,7 +501,7 @@ A key with no verdict at all is refused. Malformed input — a wrong JSON type,
 - The first row, when `system` or `developer`, becomes `Request.system`; a later `system` / `developer` row becomes a `developer` message at that position.
 - Consecutive `tool` rows become **one** tool message with one `ToolResultPart` per row (`tool_call_id` → `id`, `name` → `name`).
 - An assistant row's parts come out in a fixed order: `reasoning_content` as a `ThinkingPart`, then `content`, then `refusal`, then `tool_calls` with `arguments` parsed. `content: null` with nothing else is one empty text part (a message is never empty).
-- `text` → `TextPart`; `image_url` → `ImagePart` (a data URI becomes inline data; a URL stays a URL); `input_audio` → `AudioPart`; `file` → `DocumentPart`; `refusal` → `RefusalPart`.
+- `text` → `TextPart`; `image_url` → `ImagePart` (a data URI becomes inline data; a URL stays a URL); `input_audio` → `AudioPart` (its `format` read as the media type: `wav`, `mp3`/`mpeg`, `ogg`, `opus`, `flac`, `aac`, `aiff`, `webm`; the send raises where the wire cannot carry audio); `file` → `DocumentPart`; `refusal` → `RefusalPart`.
 - A `prompt_cache_breakpoint` on the system row is `prefix="stable"`; on the last text block of message *N* it is `prefix_until_index=N`.
 
 ### Where the round trip is not exact

lm15/providers/openai_chat.py

@@ -246,7 +246,13 @@ def _response_format_to_chat(format_config: dict[str, Any]) -> dict[str, Any]:
 
 _INGEST_GROQ_BUILTIN_INVERSE: dict[str, str] = {wire: name for name, wire in _GROQ_BUILTIN_MAP.items()}
 
-_INGEST_AUDIO_MEDIA_TYPES: dict[str, str] = {"wav": "audio/wav", "mp3": "audio/mpeg"}
+# MAP-12 rule 4: OpenAI's server takes wav and mp3, Gemini's any audio type,
+# and DSPy writes the MIME subtype (mpeg for .mp3). Each format reads as its
+# true media type; a builder with no audio slot raises at send (MAP-10).
+_INGEST_AUDIO_MEDIA_TYPES: dict[str, str] = {
+    "wav": "audio/wav", "mp3": "audio/mpeg", "mpeg": "audio/mpeg", "ogg": "audio/ogg", "opus": "audio/opus",
+    "flac": "audio/flac", "aac": "audio/aac", "aiff": "audio/aiff", "webm": "audio/webm",
+}
 
 
 def _ingest_unsupported(provider: str, what: str, why: str) -> UnsupportedFeatureError:

tests/test_openai_chat_ingest.py

@@ -19,9 +19,10 @@
 from lm15 import serde
 from lm15.compat import OpenAIChatCompat
 from lm15.errors import UnsupportedFeatureError
-from lm15.providers import OpenAIChatLM
+from lm15.providers import GeminiLM, OpenAIChatLM
 from lm15.providers.openai_chat import request_from_openai_chat
 from lm15.types import (
+    AudioPart,
     CacheConfig,
     Config,
     FunctionTool,
@@ -241,6 +242,7 @@ def test_breakpoints_map_to_cache_config() -> None:
     ({"model": "m", "messages": USER, "top_logprobs": 3}, ValueError),
     ({"model": "m", "messages": USER, "tool_choice": {"type": "function", "function": {"name": "ghost"}}}, ValueError),  # INV-031
     ({"model": "m", "messages": [{"role": "user", "content": [{"type": "image_url", "image_url": {"url": "data:image/png,notbase64"}}]}]}, ValueError),
+    ({"model": "m", "messages": [{"role": "user", "content": [{"type": "input_audio", "input_audio": {"data": "QUJD", "format": "midi"}}]}]}, ValueError),
     ({"model": "m", "messages": [{"role": "user", "content": [{"type": "text", "text": "a", "prompt_cache_breakpoint": {"mode": "explicit"}}, {"type": "text", "text": "b"}]}]}, ValueError),  # not last
     ([], TypeError),
 ])
@@ -258,6 +260,33 @@ def test_vet_op_refuses_a_non_chat_provider() -> None:
     assert out == {"canonical_request": {"model": "gpt-5-mini", "messages": [{"role": "user", "parts": [{"type": "text", "text": "Hi"}]}]}}
 
 
+# ─── input_audio formats (rule 4) ────────────────────────────────────
+
+def _audio_body(fmt: str) -> dict:
+    return {"model": "gemini-3.8-flash", "messages": [{"role": "user", "content": [
+        {"type": "text", "text": "Transcribe."},
+        {"type": "input_audio", "input_audio": {"data": "T2dnUw==", "format": fmt}},
+    ]}]}
+
+
+@pytest.mark.parametrize("fmt, media_type", [
+    ("wav", "audio/wav"), ("mp3", "audio/mpeg"), ("mpeg", "audio/mpeg"), ("ogg", "audio/ogg"),
+    ("opus", "audio/opus"), ("flac", "audio/flac"), ("aac", "audio/aac"), ("aiff", "audio/aiff"), ("webm", "audio/webm"),
+])
+def test_input_audio_reads_its_true_media_type(fmt, media_type) -> None:
+    assert request_from_openai_chat(_audio_body(fmt)) == Request(model="gemini-3.8-flash", messages=(
+        Message.user((TextPart("Transcribe."), AudioPart(media_type=media_type, data="T2dnUw=="))),
+    ))
+
+
+def test_ogg_audio_reaches_gemini_inline_and_the_chat_wire_refuses_it() -> None:
+    req = request_from_openai_chat(_audio_body("ogg"))
+    sent = json.loads(GeminiLM(api_key="k").build_request(req, stream=False).body)
+    assert sent["contents"][0]["parts"][1] == {"inlineData": {"mimeType": "audio/ogg", "data": "T2dnUw=="}}
+    with pytest.raises(UnsupportedFeatureError):
+        OpenAIChatLM(api_key="k").build_request(req, stream=False)
+
+
 # ─── message objects dumped back into history (MAP-12 addendum) ──────
 
 SDK_MESSAGE = {"content": "Ok! How can I help?", "refusal": None, "role": "assistant", "annotations": [], "audio": None, "function_call": None, "tool_calls": None}