Build with AI models that can transcribe and understand audio
With a single API call, get access to AI models built on the latest AI breakthroughs to transcribe and understand audio and speech data securely at large scale.
Visit our AssemblyAI API Documentation to get an overview of our models!
pip install -U assemblyaiBefore starting, you need to set the API key. If you don't have one yet, sign up for one!
import assemblyai as aai
# set the API key
aai.settings.api_key = f"{ASSEMBLYAI_API_KEY}"Transcribe a Local Audio File
import assemblyai as aai
transcriber = aai.Transcriber()
transcript = transcriber.transcribe("./my-local-audio-file.wav")
print(transcript.text)Transcribe an URL
import assemblyai as aai
transcriber = aai.Transcriber()
transcript = transcriber.transcribe("https://example.org/audio.mp3")
print(transcript.text)Export Subtitles of an Audio File
import assemblyai as aai
transcriber = aai.Transcriber()
transcript = transcriber.transcribe("https://example.org/audio.mp3")
# in SRT format
print(transcript.export_subtitles_srt())
# in VTT format
print(transcript.export_subtitles_vtt())List all Sentences and Paragraphs
import assemblyai as aai
transcriber = aai.Transcriber()
transcript = transcriber.transcribe("https://example.org/audio.mp3")
sentences = transcript.get_sentences()
for sentence in sentences:
print(sentence.text)
paragraphs = transcript.get_paragraphs()
for paragraph in paragraphs:
print(paragraph.text)Search for Words in a Transcript
import assemblyai as aai
transcriber = aai.Transcriber()
transcript = transcriber.transcribe("https://example.org/audio.mp3")
matches = transcript.word_search(["price", "product"])
for match in matches:
print(f"Found '{match.text}' {match.count} times in the transcript")Add Custom Spellings on a Transcript
import assemblyai as aai
config = aai.TranscriptionConfig()
config.set_custom_spelling(
{
"Kubernetes": ["k8s"],
"SQL": ["Sequel"],
}
)
transcriber = aai.Transcriber()
transcript = transcriber.transcribe("https://example.org/audio.mp3", config)
print(transcript.text)Use LeMUR to Summarize Multiple Transcripts
import assemblyai as aai
transcriber = aai.Transcriber()
transcript_group = transcriber.transcribe_group(
[
"https://example.org/customer1.mp3",
"https://example.org/customer2.mp3",
],
)
result = transcript_group.lemur.summarize(
context="Customers asking for cars",
answer_format="TLDR"
)
print(result.response)Use LeMUR to Ask Questions on a Single Transcript
import assemblyai as aai
transcriber = aai.Transcriber()
transcript = transcriber.transcribe("https://example.org/customer.mp3")
# ask some questions
questions = [
aai.LemurQuestion(question="What car was the customer interested in?"),
aai.LemurQuestion(question="What price range is the customer looking for?"),
]
result = transcript.lemur.question(questions)
for q in result.response:
print(f"Question: {q.question}")
print(f"Answer: {q.answer}")Use LeMUR to Ask for Action Items from a Single Transcript
import assemblyai as aai
transcriber = aai.Transcriber()
transcript = transcriber.transcribe("https://example.org/customer.mp3")
result = transcript.lemur.action_items(
context="Customers asking for help with resolving their problem",
answer_format="Three bullet points",
)
print(result.response)Use LeMUR to Ask Anything with a Custom Prompt
import assemblyai as aai
transcriber = aai.Transcriber()
transcript = transcriber.transcribe("https://example.org/customer.mp3")
result = transcript.lemur.task(
"You are a helpful coach. Provide an analysis of the transcript "
"and offer areas to improve with exact quotes. Include no preamble. "
"Start with an overall summary then get into the examples with feedback.",
)
print(result.response)PII Redact a Transcript
import assemblyai as aai
config = aai.TranscriptionConfig()
config.set_redact_pii(
# What should be redacted
policies=[
aai.PIIRedactionPolicy.credit_card_number,
aai.PIIRedactionPolicy.email_address,
aai.PIIRedactionPolicy.location,
aai.PIIRedactionPolicy.person_name,
aai.PIIRedactionPolicy.phone_number,
],
# How it should be redacted
substitution=aai.PIISubstitutionPolicy.hash,
)
transcriber = aai.Transcriber()
transcript = transcriber.transcribe("https://example.org/audio.mp3", config)To request a copy of the original audio file with the redacted information "beeped" out, set redact_pii_audio=True in the config.
Once the Transcript object is returned, you can access the URL of the redacted audio file with get_redacted_audio_url, or save the redacted audio directly to disk with save_redacted_audio.
import assemblyai as aai
transcript = aai.Transcriber().transcribe(
"https://example.org/audio.mp3",
config=aai.TranscriptionConfig(
redact_pii=True,
redact_pii_policies=[aai.PIIRedactionPolicy.person_name],
redact_pii_audio=True
)
)
redacted_audio_url = transcript.get_redacted_audio_url()
transcript.save_redacted_audio("redacted_audio.mp3")Summarize the content of a transcript over time
import assemblyai as aai
transcriber = aai.Transcriber()
transcript = transcriber.transcribe(
"https://example.org/audio.mp3",
config=aai.TranscriptionConfig(auto_chapters=True)
)
for chapter in transcript.chapters:
print(f"Summary: {chapter.summary}") # A one paragraph summary of the content spoken during this timeframe
print(f"Start: {chapter.start}, End: {chapter.end}") # Timestamps (in milliseconds) of the chapter
print(f"Healine: {chapter.headline}") # A single sentence summary of the content spoken during this timeframe
print(f"Gist: {chapter.gist}") # An ultra-short summary, just a few words, of the content spoken during this timeframeSummarize the content of a transcript
import assemblyai as aai
transcriber = aai.Transcriber()
transcript = transcriber.transcribe(
"https://example.org/audio.mp3",
config=aai.TranscriptionConfig(summarization=True)
)
print(transcript.summary)By default, the summarization model will be informative and the summarization type will be bullets. Read more about summarization models and types here.
To change the model and/or type, pass additional parameters to the TranscriptionConfig:
config=aai.TranscriptionConfig(
summarization=True,
summary_model=aai.SummarizationModel.catchy,
summary_type=aai.SummarizationType.headline
)Detect Sensitive Content in a Transcript
import assemblyai as aai
transcriber = aai.Transcriber()
transcript = transcriber.transcribe(
"https://example.org/audio.mp3",
config=aai.TranscriptionConfig(content_safety=True)
)
# Get the parts of the transcript which were flagged as sensitive
for result in transcript.content_safety.results:
print(result.text) # sensitive text snippet
print(result.timestamp.start)
print(result.timestamp.end)
for label in result.labels:
print(label.label) # content safety category
print(label.confidence) # model's confidence that the text is in this category
print(label.severity) # severity of the text in relation to the category
# Get the confidence of the most common labels in relation to the entire audio file
for label, confidence in transcript.content_safety.summary.items():
print(f"{confidence * 100}% confident that the audio contains {label}")
# Get the overall severity of the most common labels in relation to the entire audio file
for label, severity_confidence in transcript.content_safety.severity_score_summary.items():
print(f"{severity_confidence.low * 100}% confident that the audio contains low-severity {label}")
print(f"{severity_confidence.medium * 100}% confident that the audio contains mid-severity {label}")
print(f"{severity_confidence.high * 100}% confident that the audio contains high-severity {label}")Read more about the content safety categories.
By default, the content safety model will only include labels with a confidence greater than 0.5 (50%). To change this, pass content_safety_confidence (as an integer percentage between 25 and 100, inclusive) to the TranscriptionConfig:
config=aai.TranscriptionConfig(
content_safety=True,
content_safety_confidence=80, # only include labels with a confidence greater than 80%
)Analyze the Sentiment of Sentences in a Transcript
import assemblyai as aai
transcriber = aai.Transcriber()
transcript = transcriber.transcribe(
"https://example.org/audio.mp3",
config=aai.TranscriptionConfig(sentiment_analysis=True)
)
for sentiment_result in transcript.sentiment_analysis:
print(sentiment_result.text)
print(sentiment_result.sentiment) # POSITIVE, NEUTRAL, or NEGATIVE
print(sentiment_result.confidence)
print(f"Timestamp: {sentiment_result.timestamp.start} - {sentiment_result.timestamp.end}")If speaker_labels is also enabled, then each sentiment analysis result will also include a speaker field.
# ...
config = aai.TranscriptionConfig(sentiment_analysis=True, speaker_labels=True)
# ...
for sentiment_result in transcript.sentiment_analysis:
print(sentiment_result.speaker)Identify Entities in a Transcript
import assemblyai as aai
transcriber = aai.Transcriber()
transcript = transcriber.transcribe(
"https://example.org/audio.mp3",
config=aai.TranscriptionConfig(entity_detection=True)
)
for entity in transcript.entities:
print(entity.text) # i.e. "Dan Gilbert"
print(entity.type) # i.e. EntityType.person
print(f"Timestamp: {entity.start} - {entity.end}")Detect Topics in a Transcript (IAB Classification)
import assemblyai as aai
transcriber = aai.Transcriber()
transcript = transcriber.transcribe(
"https://example.org/audio.mp3",
config=aai.TranscriptionConfig(iab_categories=True)
)
# Get the parts of the transcript that were tagged with topics
for result in transcript.iab_categories.results:
print(result.text)
print(f"Timestamp: {result.timestamp.start} - {result.timestamp.end}")
for label in result.labels:
print(label.label) # topic
print(label.relevance) # how relevant the label is for the portion of text
# Get a summary of all topics in the transcript
for label, relevance in transcript.iab_categories.summary.items():
print(f"Audio is {relevance * 100}% relevant to {label}")Identify Important Words and Phrases in a Transcript
import assemblyai as aai
transcriber = aai.Transcriber()
transcript = transcriber.transcribe(
"https://example.org/audio.mp3",
config=aai.TranscriptionConfig(auto_highlights=True)
)
for result in transcript.auto_highlights.results:
print(result.text) # the important phrase
print(result.rank) # relevancy of the phrase
print(result.count) # number of instances of the phrase
for timestamp in result.timestamps:
print(f"Timestamp: {timestamp.start} - {timestamp.end}")Read more about our Real-Time service.
Stream your Microphone in Real-Time
import assemblyai as aai
def on_open(session_opened: aai.RealtimeSessionOpened):
"This function is called when the connection has been established."
print("Session ID:", session_opened.session_id)
def on_data(transcript: aai.RealtimeTranscript):
"This function is called when a new transcript has been received."
if not transcript.text:
return
if isinstance(transcript, aai.RealtimeFinalTranscript):
print(transcript.text, end="\r\n")
else:
print(transcript.text, end="\r")
def on_error(error: aai.RealtimeError):
"This function is called when the connection has been closed."
print("An error occured:", error)
def on_close():
"This function is called when the connection has been closed."
print("Closing Session")
# Create the Real-Time transcriber
transcriber = aai.RealtimeTranscriber(
on_data=on_data,
on_error=on_error,
sample_rate=44_100,
on_open=on_open, # optional
on_close=on_close, # optional
)
# Start the connection
transcriber.connect()
# Open a microphone stream
microphone_stream = aai.extras.MicrophoneStream()
# Press CTRL+C to abort
transcriber.stream(microphone_stream)
transcriber.close()Transcribe a Local Audio File in Real-Time
import assemblyai as aai
def on_data(transcript: aai.RealtimeTranscript):
"This function is called when a new transcript has been received."
if not transcript.text:
return
if isinstance(transcript, aai.RealtimeFinalTranscript):
print(transcript.text, end="\r\n")
else:
print(transcript.text, end="\r")
def on_error(error: aai.RealtimeError):
"This function is called when the connection has been closed."
print("An error occured:", error)
# Create the Real-Time transcriber
transcriber = aai.RealtimeTranscriber(
on_data=on_data,
on_error=on_error,
sample_rate=44_100,
)
# Start the connection
transcriber.connect()
# Only WAV/PCM16 single channel supported for now
file_stream = aai.extras.stream_file(
filepath="audio.wav",
sample_rate=44_100,
)
transcriber.stream(file_stream)
transcriber.close()Visit one of our Playgrounds:
When no TranscriptionConfig is being passed to the Transcriber or its methods, it will use a default instance of a TranscriptionConfig.
If you would like to re-use the same TranscriptionConfig for all your transcriptions,
you can set it on the Transcriber directly:
config = aai.TranscriptionConfig(punctuate=False, format_text=False)
transcriber = aai.Transcriber(config=config)
# will use the same config for all `.transcribe*(...)` operations
transcriber.transcribe("https://example.org/audio.wav")You can override the default configuration later via the .config property of the Transcriber:
transcriber = aai.Transcriber()
# override the `Transcriber`'s config with a new config
transcriber.config = aai.TranscriptionConfig(punctuate=False, format_text=False)In case you want to override the Transcriber's configuration for a specific operation with a different one, you can do so via the config parameter of a .transcribe*(...) method:
config = aai.TranscriptionConfig(punctuate=False, format_text=False)
# set a default configuration
transcriber = aai.Transcriber(config=config)
transcriber.transcribe(
"https://example.com/audio.mp3",
# overrides the above configuration on the `Transcriber` with the following
config=aai.TranscriptionConfig(dual_channel=True, disfluencies=True)
)Currently, the SDK provides two ways to transcribe audio files.
The synchronous approach halts the application's flow until the transcription has been completed.
The asynchronous approach allows the application to continue running while the transcription is being processed. The caller receives a concurrent.futures.Future object which can be used to check the status of the transcription at a later time.
You can identify those two approaches by the _async suffix in the Transcriber's method name (e.g. transcribe vs transcribe_async).
By default we poll the Transcript's status each 3s. In case you would like to adjust that interval:
import assemblyai as aai
aai.settings.polling_interval = 1.0If you previously created a transcript, you can use its ID to retrieve it later.
import assemblyai as aai
transcript = aai.Transcript.get_by_id("<TRANSCRIPT_ID>")
print(transcript.id)
print(transcript.text)You can also retrieve multiple existing transcripts and combine them into a single TranscriptGroup object. This allows you to perform operations on the transcript group as a single unit, such as querying the combined transcripts with LeMUR.
import assemblyai as aai
transcript_group = aai.TranscriptGroup.get_by_ids(["<TRANSCRIPT_ID_1>", "<TRANSCRIPT_ID_2>"])
summary = transcript_group.lemur.summarize(context="Customers asking for cars", answer_format="TLDR")
print(summary)Both Transcript.get_by_id and TranscriptGroup.get_by_ids have asynchronous counterparts, Transcript.get_by_id_async and TranscriptGroup.get_by_ids_async, respectively. These functions immediately return a Future object, rather than blocking until the transcript(s) are retrieved.
See the above section on Synchronous vs Asynchronous for more information.