Audio transcription support

CHANGELOG.md

@@ -9,6 +9,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ### Added
 - `AzureOpenAiStructureConfig` for providing Structures with all Azure OpenAI Driver configuration.
 - `AzureOpenAiVisionImageQueryDriver` to support queries on images using Azure's OpenAI Vision models.
+- `AudioLoader` for loading audio content into an `AudioArtifact`.
+- `AudioTranscriptionTask` and `AudioTranscriptionClient` for transcribing audio content in Structures.
+- `OpenAiAudioTranscriptionDriver` for integration with OpenAI's speech-to-text models, including Whisper.
 - Parameter `env` to `BaseStructureRunDriver` to set environment variables for a Structure Run.
 
 ### Changed

docs/griptape-framework/data/loaders.md

@@ -199,3 +199,24 @@ loader.load(EmailLoader.EmailQuery(label="INBOX"))
 
 loader.load_collection([EmailLoader.EmailQuery(label="INBOX"), EmailLoader.EmailQuery(label="SENT")])
 ```
+
+## Audio Loader 
+
+!!! info
+    This driver requires the `loaders-audio` [extra](../index.md#extras).
+
+The [Audio Loader](../../reference/griptape/loaders/audio_loader.md) is used to load audio content as an [AudioArtifact](./artifacts.md#audioartifact). The Loader operates on audio bytes that can be sourced from files on disk, downloaded audio, or audio in memory.
+
+The Loader will load audio in its native format and populates the resulting Artifact's `format` field by making a best-effort guess of the underlying audio format using the `filetype` package.
+
+```python
+from griptape.loaders import AudioLoader
+from griptape.utils import load_file
+
+# Load an image from disk
+with open("tests/resources/sentences.wav", "rb") as f:
+    audio_artifact = AudioLoader().load(f.read())
+
+# You can also use the load_file utility function
+AudioLoader().load(load_file("tests/resources/sentences.wav"))
+```

docs/griptape-framework/drivers/audio-transcription-drivers.md

@@ -0,0 +1,32 @@
+## Overview
+
+[Audio Transcription Drivers](../../reference/griptape/drivers/audio_transcription/index.md) extract text from spoken audio.
+
+This driver acts as a critical bridge between audio transcription Engines and the underlying models, facilitating the construction and execution of API calls that transform speech into editable and searchable text. Utilized predominantly in applications that support the input of verbal communications, the Audio Transcription Driver effectively extracts and interprets speech, rendering it into a textual format that can be easily integrated into data systems and Workflows.
+
+This capability is essential for enhancing accessibility, improving content discoverability, and automating tasks that traditionally relied on manual transcription, thereby streamlining operations and enhancing efficiency across various industries.
+
+### OpenAI
+
+The [OpenAI Audio Transcription Driver](../../reference/griptape/drivers/audio_transcription/openai_audio_transcription_driver.md) utilizes OpenAI's sophisticated `whisper` model to accurately transcribe spoken audio into text. This model supports multiple languages, ensuring precise transcription across a wide range of dialects. 
+
+```python
+from griptape.drivers import OpenAiAudioTranscriptionDriver
+from griptape.engines import AudioTranscriptionEngine
+from griptape.tools.audio_transcription_client.tool import AudioTranscriptionClient
+from griptape.structures import Agent
+
+
+driver = OpenAiAudioTranscriptionDriver(
+    model="whisper-1"
+)
+
+tool = AudioTranscriptionClient(
+    off_prompt=False,
+    engine=AudioTranscriptionEngine(
+        audio_transcription_driver=driver,
+    ),
+)
+
+Agent(tools=[tool]).run("Transcribe the following audio file: tests/resources/sentences.wav")
+```

docs/griptape-framework/engines/audio-engines.md

@@ -27,3 +27,26 @@ engine.run(
     prompts=["Hello, world!"],
 )
 ```
+
+### Audio Transcription Engine
+
+The [Audio Transcription Engine](../../reference/griptape/engines/audio/audio_transcription_engine.md) facilitates transcribing speech from audio inputs.
+
+```python
+from griptape.drivers import OpenAiAudioTranscriptionDriver
+from griptape.engines import AudioTranscriptionEngine
+from griptape.loaders import AudioLoader
+from griptape.utils import load_file
+
+
+driver = OpenAiAudioTranscriptionDriver(
+    model="whisper-1"
+)
+
+engine = AudioTranscriptionEngine(
+    audio_transcription_driver=driver,
+)
+
+audio_artifact = AudioLoader().load(load_file("tests/resources/sentences.wav"))
+engine.run(audio_artifact)
+```

docs/griptape-framework/structures/tasks.md

@@ -805,3 +805,57 @@ team = Pipeline(
 
 team.run()
 ```
+
+## Text to Speech Task
+
+This Task enables Structures to synthesize speech from text using [Text to Speech Engines](../../reference/griptape/engines/audio/text_to_speech_engine.md) and [Text to Speech Drivers](../../reference/griptape/drivers/text_to_speech/index.md).
+
+```python
+import os
+
+from griptape.drivers import ElevenLabsTextToSpeechDriver
+from griptape.engines import TextToSpeechEngine
+from griptape.tasks import TextToSpeechTask
+from griptape.structures import Pipeline
+
+
+driver = ElevenLabsTextToSpeechDriver(
+    api_key=os.getenv("ELEVEN_LABS_API_KEY"),
+    model="eleven_multilingual_v2",
+    voice="Matilda",
+)
+
+task = TextToSpeechTask(
+    text_to_speech_engine=TextToSpeechEngine(
+        text_to_speech_driver=driver,
+    ),
+)
+
+Pipeline(tasks=[task]).run("Generate audio from this text: 'Hello, world!'")
+```
+
+## Audio Transcription Task 
+
+This Task enables Structures to transcribe speech from text using [Audio Transcription Engines](../../reference/griptape/engines/audio/audio_transcription_engine.md) and [Audio Transcription Drivers](../../reference/griptape/drivers/audio_transcription/index.md).
+
+```python
+from griptape.drivers import OpenAiAudioTranscriptionDriver
+from griptape.engines import AudioTranscriptionEngine
+from griptape.loaders import AudioLoader
+from griptape.tasks import AudioTranscriptionTask
+from griptape.structures import Pipeline
+from griptape.utils import load_file
+
+driver = OpenAiAudioTranscriptionDriver(
+    model="whisper-1"
+)
+
+task = AudioTranscriptionTask(
+    input=lambda _: AudioLoader().load(load_file("tests/resources/sentences2.wav")),
+    audio_transcription_engine=AudioTranscriptionEngine(
+        audio_transcription_driver=driver,
+    ),
+)
+
+Pipeline(tasks=[task]).run()
+```

docs/griptape-tools/official-tools/audio-transcription-client.md

@@ -0,0 +1,24 @@
+# AudioTranscriptionClient
+
+This Tool enables [Agents](../../griptape-framework/structures/agents.md) to transcribe speech from text using [Audio Transcription Engines](../../reference/griptape/engines/audio/audio_transcription_engine.md) and [Audio Transcription Drivers](../../reference/griptape/drivers/audio_transcription/index.md).
+
+```python
+from griptape.drivers import OpenAiAudioTranscriptionDriver
+from griptape.engines import AudioTranscriptionEngine
+from griptape.tools.audio_transcription_client.tool import AudioTranscriptionClient
+from griptape.structures import Agent
+
+
+driver = OpenAiAudioTranscriptionDriver(
+    model="whisper-1"
+)
+
+tool = AudioTranscriptionClient(
+    off_prompt=False,
+    engine=AudioTranscriptionEngine(
+        audio_transcription_driver=driver,
+    ),
+)
+
+Agent(tools=[tool]).run("Transcribe the following audio file: /Users/andrew/code/griptape/tests/resources/sentences2.wav")
+```

docs/griptape-tools/official-tools/text-to-speech-client.md

@@ -1,6 +1,6 @@
 # TextToSpeechClient
 
-This tool enables LLMs to synthesize speech from text using [Text to Speech Engines](../../reference/griptape/engines/audio/text_to_speech_engine.md) and [Text to Speech Drivers](../../reference/griptape/drivers/text_to_speech/index.md).
+This Tool enables LLMs to synthesize speech from text using [Text to Speech Engines](../../reference/griptape/engines/audio/text_to_speech_engine.md) and [Text to Speech Drivers](../../reference/griptape/drivers/text_to_speech/index.md).
 
 ```python
 import os

griptape/config/base_structure_config.py

@@ -14,6 +14,7 @@
     BasePromptDriver,
     BaseVectorStoreDriver,
     BaseTextToSpeechDriver,
+    BaseAudioTranscriptionDriver,
 )
 from griptape.utils import dict_merge
 
@@ -29,6 +30,7 @@ class BaseStructureConfig(BaseConfig, ABC):
         default=None, kw_only=True, metadata={"serializable": True}
     )
     text_to_speech_driver: BaseTextToSpeechDriver = field(kw_only=True, metadata={"serializable": True})
+    audio_transcription_driver: BaseAudioTranscriptionDriver = field(kw_only=True, metadata={"serializable": True})
 
     def merge_config(self, config: dict) -> BaseStructureConfig:
         base_config = self.to_dict()

griptape/config/openai_structure_config.py

@@ -11,6 +11,10 @@
     OpenAiChatPromptDriver,
     OpenAiEmbeddingDriver,
     OpenAiImageGenerationDriver,
+    BaseTextToSpeechDriver,
+    OpenAiTextToSpeechDriver,
+    BaseAudioTranscriptionDriver,
+    OpenAiAudioTranscriptionDriver,
     OpenAiImageQueryDriver,
 )
 
@@ -40,3 +44,11 @@ class OpenAiStructureConfig(StructureConfig):
         kw_only=True,
         metadata={"serializable": True},
     )
+    text_to_speech_driver: BaseTextToSpeechDriver = field(
+        default=Factory(lambda: OpenAiTextToSpeechDriver(model="tts")), kw_only=True, metadata={"serializable": True}
+    )
+    audio_transcription_driver: BaseAudioTranscriptionDriver = field(
+        default=Factory(lambda: OpenAiAudioTranscriptionDriver(model="whisper-1")),
+        kw_only=True,
+        metadata={"serializable": True},
+    )

griptape/config/structure_config.py

@@ -17,6 +17,8 @@
     BaseImageQueryDriver,
     BaseTextToSpeechDriver,
     DummyTextToSpeechDriver,
+    BaseAudioTranscriptionDriver,
+    DummyAudioTranscriptionDriver,
 )
 
 
@@ -43,3 +45,6 @@ class StructureConfig(BaseStructureConfig):
     text_to_speech_driver: BaseTextToSpeechDriver = field(
         default=Factory(lambda: DummyTextToSpeechDriver()), kw_only=True, metadata={"serializable": True}
     )
+    audio_transcription_driver: BaseAudioTranscriptionDriver = field(
+        default=Factory(lambda: DummyAudioTranscriptionDriver()), kw_only=True, metadata={"serializable": True}
+    )

griptape/drivers/__init__.py

@@ -107,6 +107,10 @@
 from .structure_run.griptape_cloud_structure_run_driver import GriptapeCloudStructureRunDriver
 from .structure_run.local_structure_run_driver import LocalStructureRunDriver
 
+from .audio_transcription.base_audio_transcription_driver import BaseAudioTranscriptionDriver
+from .audio_transcription.dummy_audio_transcription_driver import DummyAudioTranscriptionDriver
+from .audio_transcription.openai_audio_transcription_driver import OpenAiAudioTranscriptionDriver
+
 __all__ = [
     "BasePromptDriver",
     "OpenAiChatPromptDriver",
@@ -199,4 +203,7 @@
     "BaseStructureRunDriver",
     "GriptapeCloudStructureRunDriver",
     "LocalStructureRunDriver",
+    "BaseAudioTranscriptionDriver",
+    "DummyAudioTranscriptionDriver",
+    "OpenAiAudioTranscriptionDriver",
 ]

griptape/drivers/audio_transcription/base_audio_transcription_driver.py

@@ -0,0 +1,42 @@
+from __future__ import annotations
+
+from abc import ABC, abstractmethod
+from typing import TYPE_CHECKING, Optional
+
+from attrs import define, field
+
+from griptape.artifacts import TextArtifact, AudioArtifact
+from griptape.events import StartAudioTranscriptionEvent, FinishAudioTranscriptionEvent
+from griptape.mixins import ExponentialBackoffMixin, SerializableMixin
+
+if TYPE_CHECKING:
+    from griptape.structures import Structure
+
+
+@define
+class BaseAudioTranscriptionDriver(SerializableMixin, ExponentialBackoffMixin, ABC):
+    model: str = field(kw_only=True, metadata={"serializable": True})
+    structure: Optional[Structure] = field(default=None, kw_only=True)
+
+    def before_run(self) -> None:
+        if self.structure:
+            self.structure.publish_event(StartAudioTranscriptionEvent())
+
+    def after_run(self) -> None:
+        if self.structure:
+            self.structure.publish_event(FinishAudioTranscriptionEvent())
+
+    def run(self, audio: AudioArtifact, prompts: Optional[list[str]] = None) -> TextArtifact:
+        for attempt in self.retrying():
+            with attempt:
+                self.before_run()
+                result = self.try_run(audio, prompts)
+                self.after_run()
+
+                return result
+
+        else:
+            raise Exception("Failed to run audio transcription")
+
+    @abstractmethod
+    def try_run(self, audio: AudioArtifact, prompts: Optional[list[str]] = None) -> TextArtifact: ...

griptape/drivers/audio_transcription/dummy_audio_transcription_driver.py

@@ -0,0 +1,14 @@
+from typing import Optional
+
+from attrs import define, field
+from griptape.artifacts import AudioArtifact, TextArtifact
+from griptape.drivers import BaseAudioTranscriptionDriver
+from griptape.exceptions import DummyException
+
+
+@define
+class DummyAudioTranscriptionDriver(BaseAudioTranscriptionDriver):
+    model: str = field(init=False)
+
+    def try_run(self, audio: AudioArtifact, prompts: Optional[list] = None) -> TextArtifact:
+        raise DummyException(__class__.__name__, "try_transcription")

griptape/drivers/audio_transcription/openai_audio_transcription_driver.py

@@ -0,0 +1,43 @@
+from __future__ import annotations
+
+import io
+from typing import Optional
+
+import openai
+from attrs import field, Factory, define
+
+from griptape.artifacts import AudioArtifact, TextArtifact
+from griptape.drivers import BaseAudioTranscriptionDriver
+
+
+@define
+class OpenAiAudioTranscriptionDriver(BaseAudioTranscriptionDriver):
+    api_type: str = field(default=openai.api_type, kw_only=True)
+    api_version: Optional[str] = field(default=openai.api_version, kw_only=True, metadata={"serializable": True})
+    base_url: Optional[str] = field(default=None, kw_only=True, metadata={"serializable": True})
+    api_key: Optional[str] = field(default=None, kw_only=True, metadata={"serializable": False})
+    organization: Optional[str] = field(default=openai.organization, kw_only=True, metadata={"serializable": True})
+    client: openai.OpenAI = field(
+        default=Factory(
+            lambda self: openai.OpenAI(api_key=self.api_key, base_url=self.base_url, organization=self.organization),
+            takes_self=True,
+        )
+    )
+
+    def try_run(self, audio: AudioArtifact, prompts: Optional[list[str]] = None) -> TextArtifact:
+        additional_params = {}
+
+        if prompts is not None:
+            additional_params["prompt"] = ", ".join(prompts)
+
+        transcription = self.client.audio.transcriptions.create(
+            # Even though we're not actually providing a file to the client, the API still requires that we send a file
+            # name. We set the file name to use the same format as the audio file so that the API can reject
+            # it if the format is unsupported.
+            model=self.model,
+            file=(f"a.{audio.format}", io.BytesIO(audio.value)),
+            response_format="json",
+            **additional_params,
+        )
+
+        return TextArtifact(value=transcription.text)

griptape/engines/__init__.py

@@ -12,6 +12,7 @@
 from .image.outpainting_image_generation_engine import OutpaintingImageGenerationEngine
 from .image_query.image_query_engine import ImageQueryEngine
 from .audio.text_to_speech_engine import TextToSpeechEngine
+from .audio.audio_transcription_engine import AudioTranscriptionEngine
 
 __all__ = [
     "BaseQueryEngine",
@@ -28,4 +29,5 @@
     "OutpaintingImageGenerationEngine",
     "ImageQueryEngine",
     "TextToSpeechEngine",
+    "AudioTranscriptionEngine",
 ]

griptape/engines/audio/audio_transcription_engine.py

@@ -0,0 +1,12 @@
+from attrs import define, field
+
+from griptape.artifacts import AudioArtifact, TextArtifact
+from griptape.drivers import BaseAudioTranscriptionDriver
+
+
+@define
+class AudioTranscriptionEngine:
+    audio_transcription_driver: BaseAudioTranscriptionDriver = field(kw_only=True)
+
+    def run(self, audio: AudioArtifact, *args, **kwargs) -> TextArtifact:
+        return self.audio_transcription_driver.try_run(audio)